instance_id stringlengths 37 102 | patch stringclasses 0
values | FAIL_TO_PASS listlengths | PASS_TO_PASS listlengths | image_name stringclasses 0
values | repo stringclasses 0
values | problem_statement stringlengths 274 9.46k | task_id stringlengths 8 42 | domain stringclasses 0
values | system_prompt stringclasses 0
values | user_prompt stringclasses 0
values | selected_tools listlengths | restricted_tools listlengths | mcp_endpoint stringclasses 0
values | number_of_runs int64 | reset_database_between_runs bool 0
classes | gym_servers_config stringclasses 0
values | verifiers stringclasses 0
values | benchmark stringclasses 1
value | enterprise_ops_test_result stringclasses 0
values | skillsbench_test_result stringlengths 1.42k 109k | instruction stringlengths 274 9.46k | num_actions int64 6 1.74k | conversation_archive_path stringlengths 147 227 | benchmark_run_name stringclasses 3
values | resolved bool 2
classes | model_pred_patch stringclasses 0
values | model stringclasses 3
values | eval_report stringclasses 0
values | eval_test_output stringclasses 0
values | prompt_tokens int64 66.5k 12.1M | reasoning_tokens int64 646 244k | completion_tokens int64 2.28k 284k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
skills-gemini-gemini-3.5-flash-oh-crystallographic-wyckoff-position-analysis | null | null | null | null | null | You are a materials science researcher. You want to analyze the crystal structure of a few samples. You are given a few CIF files generated by SHELX through X-ray Diffraction.
To process the given CIF file data in batches, you are required to build a function in python to analyze Wyckoff position multiplicities and ca... | crystallographic-wyckoff-position-analysis | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/crystallographic-wyckoff-position-analysis","passed":false,"reward":0.45,"task_image":... | You are a materials science researcher. You want to analyze the crystal structure of a few samples. You are given a few CIF files generated by SHELX through X-ray Diffraction.
To process the given CIF file data in batches, you are required to build a function in python to analyze Wyckoff position multiplicities and ca... | 31 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/crystallographic-wyckoff-position-analysis.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 980,999 | 20,108 | 26,452 |
skills-gpt-5.5-2026-04-23-oh-ada-bathroom-plan-repair | null | null | null | null | null | # Task
You are given a standardized 2D toilet-room CAD plan. Extract the original room layout, identify the plan-view ADA (Americans with Disabilities Act)-derived accessibility violations, and produce a minimally invasive repaired CAD layout that satisfies the provided rules while preserving the original design where... | ada-bathroom-plan-repair | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/ada-bathroom-plan-repair","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/ada-bathroo... | # Task
You are given a standardized 2D toilet-room CAD plan. Extract the original room layout, identify the plan-view ADA (Americans with Disabilities Act)-derived accessibility violations, and produce a minimally invasive repaired CAD layout that satisfies the provided rules while preserving the original design where... | 16 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/ada-bathroom-plan-repair.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 588,392 | 11,119 | 26,684 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-shock-analysis-supply | null | null | null | null | null | This task is to estimate an investment spending shock to a small open economy (Georgia) using Cobb-Douglas production function to estimate the potential GDP. The investment will last 8 years beginning from 2026, which worths 6.5 billion USD. You are going to collect the relevant data from multiple online sources, popul... | shock-analysis-supply | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/shock-analysis-supply/critic_attempt_1/run_c3341ced34e... | This task is to estimate an investment spending shock to a small open economy (Georgia) using Cobb-Douglas production function to estimate the potential GDP. The investment will last 8 years beginning from 2026, which worths 6.5 billion USD. You are going to collect the relevant data from multiple online sources, popul... | 86 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/shock-analysis-supply.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 5,585,891 | 172,864 | 215,790 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-lake-warming-attribution | null | null | null | null | null | My data is in `/root/data/`, which includes:
1. `water_temperature.csv`: Lake surface temperature (0-5m)
2. `climate.csv`: Climate variables
3. `land_cover.csv`: Land cover data
4. `hydrology.csv`: Hydrology data
First, I want to do a trend analysis to determine whether there is a long-term warming trend for the water... | lake-warming-attribution | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/lake-warming-attribution/critic_attempt_1/run_6275d07b... | My data is in `/root/data/`, which includes:
1. `water_temperature.csv`: Lake surface temperature (0-5m)
2. `climate.csv`: Climate variables
3. `land_cover.csv`: Land cover data
4. `hydrology.csv`: Hydrology data
First, I want to do a trend analysis to determine whether there is a long-term warming trend for the water... | 12 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/lake-warming-attribution.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 134,145 | 2,924 | 6,655 |
skills-gpt-5.5-2026-04-23-oh-fix-erlang-ssh-cve | null | null | null | null | null | Erlang/OTP SSH is the built-in SSH server component of the Erlang/OTP platform.
A critical vulnerability was discovered in the Erlang/OTP SSH server, allowing attackers to execute arbitrary system commands remotely without authentication by crafting specific SSH protocol messages. In `/app/workspace/otp_src_27.3.2` we... | fix-erlang-ssh-cve | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/fix-erlang-ssh-cve","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/fix-erlang-ssh-cv... | Erlang/OTP SSH is the built-in SSH server component of the Erlang/OTP platform.
A critical vulnerability was discovered in the Erlang/OTP SSH server, allowing attackers to execute arbitrary system commands remotely without authentication by crafting specific SSH protocol messages. In `/app/workspace/otp_src_27.3.2` we... | 35 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/fix-erlang-ssh-cve.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 1,284,276 | 5,511 | 10,081 |
skills-gemini-gemini-3.5-flash-oh-adaptive-cruise-control | null | null | null | null | null | You need to implement an Adaptive Cruise Control (ACC) simulation that maintains the set speed (30m/s) when no vehicles are detected ahead, and automatically adjusts speed to maintain a safe following distance when a vehicle is detected ahead. The targets are: speed rise time <10s, speed overshoot <5%, speed steady-sta... | adaptive-cruise-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/adaptive-cruise-control","passed":true,"reward":1.0,"task_image":"openhands-skillsbenc... | You need to implement an Adaptive Cruise Control (ACC) simulation that maintains the set speed (30m/s) when no vehicles are detected ahead, and automatically adjusts speed to maintain a safe following distance when a vehicle is detected ahead. The targets are: speed rise time <10s, speed overshoot <5%, speed steady-sta... | 63 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/adaptive-cruise-control.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 2,998,881 | 40,166 | 55,966 |
skills-gemini-gemini-3.5-flash-oh-flood-risk-analysis | null | null | null | null | null | Find stations experienced flooding during April 1-7, 2025. USGS stations data is in `/root/data/michigan_stations.txt` which contains streamflow records. You should output a CSV file to `/root/output/flood_results.csv`. The 'flood_results.csv' should have two columns: `station_id` (the USGS station ID) and `flood_days... | flood-risk-analysis | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/flood-risk-analysis","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/f... | Find stations experienced flooding during April 1-7, 2025. USGS stations data is in `/root/data/michigan_stations.txt` which contains streamflow records. You should output a CSV file to `/root/output/flood_results.csv`. The 'flood_results.csv' should have two columns: `station_id` (the USGS station ID) and `flood_days... | 41 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/flood-risk-analysis.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 1,301,997 | 13,831 | 23,554 |
skills-gpt-5.5-2026-04-23-oh-invoice-fraud-detection | null | null | null | null | null | You need to analyze the following files to find any potential invoice fraud:
- `/root/invoices.pdf`: One invoice per page.
- `/root/vendors.xlsx`: Approved vendors with corresponding Vendor ID, Name, and authorized IBAN.
- `/root/purchase_orders.csv`: Valid purchase orders with corresponding PO Number, Amount, and Vend... | invoice-fraud-detection | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/invoice-fraud-detection","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/invoice-fra... | You need to analyze the following files to find any potential invoice fraud:
- `/root/invoices.pdf`: One invoice per page.
- `/root/vendors.xlsx`: Approved vendors with corresponding Vendor ID, Name, and authorized IBAN.
- `/root/purchase_orders.csv`: Valid purchase orders with corresponding PO Number, Amount, and Vend... | 17 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/invoice-fraud-detection.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 297,328 | 1,816 | 6,521 |
skills-gemini-gemini-3.5-flash-oh-financial-modeling-qa | null | null | null | null | null | Please analysis large-scale data file and give me answer for a financial data analysis question.
The data file is located at /root/data.xlsx.
The background of the question is located at /root/background.pdf.
The question is: If the odd numbered games are played by Player 1, and the even numbered games are played by... | financial-modeling-qa | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/financial-modeling-qa","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/... | Please analysis large-scale data file and give me answer for a financial data analysis question.
The data file is located at /root/data.xlsx.
The background of the question is located at /root/background.pdf.
The question is: If the odd numbered games are played by Player 1, and the even numbered games are played by... | 48 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/financial-modeling-qa.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 1,645,233 | 27,561 | 38,744 |
skills-gemini-gemini-3.5-flash-oh-energy-ac-optimal-power-flow | null | null | null | null | null | As an operator-planner in an Independent System Operator, you create a base case for tomorrow's peak hour each morning in order to establish the voltage profile and verify that the day-ahead market schedule is AC-feasible. In order to do this, you should use the model described in `math-model.md` to find a least-cost, ... | energy-ac-optimal-power-flow | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/energy-ac-optimal-power-flow","passed":false,"reward":0.0,"task_image":"openhands-skil... | As an operator-planner in an Independent System Operator, you create a base case for tomorrow's peak hour each morning in order to establish the voltage profile and verify that the day-ahead market schedule is AC-feasible. In order to do this, you should use the model described in `math-model.md` to find a least-cost, ... | 53 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/energy-ac-optimal-power-flow.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 2,184,815 | 30,260 | 47,699 |
skills-gemini-gemini-3.5-flash-oh-lake-warming-attribution | null | null | null | null | null | My data is in `/root/data/`, which includes:
1. `water_temperature.csv`: Lake surface temperature (0-5m)
2. `climate.csv`: Climate variables
3. `land_cover.csv`: Land cover data
4. `hydrology.csv`: Hydrology data
First, I want to do a trend analysis to determine whether there is a long-term warming trend for the water... | lake-warming-attribution | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/lake-warming-attribution","passed":true,"reward":1.0,"task_image":"openhands-skillsben... | My data is in `/root/data/`, which includes:
1. `water_temperature.csv`: Lake surface temperature (0-5m)
2. `climate.csv`: Climate variables
3. `land_cover.csv`: Land cover data
4. `hydrology.csv`: Hydrology data
First, I want to do a trend analysis to determine whether there is a long-term warming trend for the water... | 41 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/lake-warming-attribution.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 1,205,030 | 20,992 | 31,707 |
skills-gpt-5.5-2026-04-23-oh-shock-analysis-supply | null | null | null | null | null | This task is to estimate an investment spending shock to a small open economy (Georgia) using Cobb-Douglas production function to estimate the potential GDP. The investment will last 8 years beginning from 2026, which worths 6.5 billion USD. You are going to collect the relevant data from multiple online sources, popul... | shock-analysis-supply | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/shock-analysis-supply","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/shock-analysi... | This task is to estimate an investment spending shock to a small open economy (Georgia) using Cobb-Douglas production function to estimate the potential GDP. The investment will last 8 years beginning from 2026, which worths 6.5 billion USD. You are going to collect the relevant data from multiple online sources, popul... | 60 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/shock-analysis-supply.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 3,226,528 | 17,299 | 39,284 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-edit-pdf | null | null | null | null | null | You are given a `PDF` file at `/root/input/input.pdf` and a text file at `/root/input/input.txt` containing natural language instructions and updated information.
Your task is to update this PDF using the latest information provided in the text file and follow all the instructions in it.
Use the text file as the sourc... | edit-pdf | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/edit-pdf/critic_attempt_1/run_25da3b16d8d64713a3eabf48... | You are given a `PDF` file at `/root/input/input.pdf` and a text file at `/root/input/input.txt` containing natural language instructions and updated information.
Your task is to update this PDF using the latest information provided in the text file and follow all the instructions in it.
Use the text file as the sourc... | 50 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/edit-pdf.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 1,420,771 | 59,438 | 75,091 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-reserves-at-risk-calc | null | null | null | null | null | STEP 1: download global commodity excel database from https://www.imf.org/en/research/commodity-prices, extract gold price (US$ per troy ounce) and populate the data into "Gold price" sheet. Calculate the monthly log, 3-month volatility and 12-month volatility in the relevant column (column C D E). Go to sheet "Answer"... | reserves-at-risk-calc | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/reserves-at-risk-calc/critic_attempt_1/run_27569a1b217... | STEP 1: download global commodity excel database from https://www.imf.org/en/research/commodity-prices, extract gold price (US$ per troy ounce) and populate the data into "Gold price" sheet. Calculate the monthly log, 3-month volatility and 12-month volatility in the relevant column (column C D E). Go to sheet "Answer"... | 58 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/reserves-at-risk-calc.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 2,043,256 | 63,140 | 85,635 |
skills-gemini-gemini-3.5-flash-oh-ada-bathroom-plan-repair | null | null | null | null | null | # Task
You are given a standardized 2D toilet-room CAD plan. Extract the original room layout, identify the plan-view ADA (Americans with Disabilities Act)-derived accessibility violations, and produce a minimally invasive repaired CAD layout that satisfies the provided rules while preserving the original design where... | ada-bathroom-plan-repair | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/ada-bathroom-plan-repair","passed":false,"reward":0.0,"task_image":"openhands-skillsbe... | # Task
You are given a standardized 2D toilet-room CAD plan. Extract the original room layout, identify the plan-view ADA (Americans with Disabilities Act)-derived accessibility violations, and produce a minimally invasive repaired CAD layout that satisfies the provided rules while preserving the original design where... | 57 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/ada-bathroom-plan-repair.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 3,142,672 | 37,040 | 54,709 |
skills-gpt-5.5-2026-04-23-oh-earthquake-plate-calculation | null | null | null | null | null | You are a geospatial analyst with expertise in plate tectonics and earthquake analysis. I want to find the earthquake that occurred furthest from the Pacific plate boundary within the Pacific plate itself. Use GeoPandas projections. Output the result to `/root/answer.json` as a JSON file with the following fields:
- `... | earthquake-plate-calculation | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/earthquake-plate-calculation","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/earthqu... | You are a geospatial analyst with expertise in plate tectonics and earthquake analysis. I want to find the earthquake that occurred furthest from the Pacific plate boundary within the Pacific plate itself. Use GeoPandas projections. Output the result to `/root/answer.json` as a JSON file with the following fields:
- `... | 13 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/earthquake-plate-calculation.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 245,023 | 5,948 | 10,384 |
skills-gemini-gemini-3.5-flash-oh-python-scala-translation | null | null | null | null | null | # Python to Scala Code Translation
'/root/Tokenizer.py' is a python code for data preparation, and you need to translate this code into Scala for processing massive data in distributed systems. You will need to make sure your Scala code follows the best practices and save your file in `/root/Tokenizer.scala`. Your Sca... | python-scala-translation | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/python-scala-translation","passed":false,"reward":0.0,"task_image":"openhands-skillsbe... | # Python to Scala Code Translation
'/root/Tokenizer.py' is a python code for data preparation, and you need to translate this code into Scala for processing massive data in distributed systems. You will need to make sure your Scala code follows the best practices and save your file in `/root/Tokenizer.scala`. Your Sca... | 29 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/python-scala-translation.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 1,292,683 | 11,543 | 15,865 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-manufacturing-codebook-normalization | null | null | null | null | null | At manufacturing test centers, testing engineers often write recognized defect reasons quickly with typos, noise, abbreviations, Chinese-English characters mixtures, etc. These texts vary largely between different testing engineers. Testing engineers are given standard codebooks but they may not follow this standardiza... | manufacturing-codebook-normalization | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/manufacturing-codebook-normalization/critic_attempt_1/... | At manufacturing test centers, testing engineers often write recognized defect reasons quickly with typos, noise, abbreviations, Chinese-English characters mixtures, etc. These texts vary largely between different testing engineers. Testing engineers are given standard codebooks but they may not follow this standardiza... | 73 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/manufacturing-codebook-normalization.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 3,961,247 | 66,031 | 97,803 |
skills-gpt-5.5-2026-04-23-oh-jpg-ocr-stat | null | null | null | null | null | ## Task description
In `/app/workspace/dataset/img`, I provide a set of scanned receipt images. Each receipt image contains text such as date, product name, unit price, total amount cost, etc. The text mainly consists of digits and English characters.
Read all image files under the given path, extract their data and ... | jpg-ocr-stat | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/jpg-ocr-stat","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/jpg-ocr-stat:b59d603fd6... | ## Task description
In `/app/workspace/dataset/img`, I provide a set of scanned receipt images. Each receipt image contains text such as date, product name, unit price, total amount cost, etc. The text mainly consists of digits and English characters.
Read all image files under the given path, extract their data and ... | 20 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/jpg-ocr-stat.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 496,196 | 1,850 | 6,558 |
skills-gpt-5.5-2026-04-23-oh-sec-financial-report | null | null | null | null | null | You are a financial analyst in hedge fund, your task is to analyze the activities of hedge funds in q3 2025 by compare it with q2 2025. The dataset is downloaded to `/root/2025-q2` and `/root/2025-q3` folders respectively.
Questions to be answered in this task:
1. In Q3, what's the AUM of Renaissance Technologies fou... | sec-financial-report | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/sec-financial-report","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/sec-financial-... | You are a financial analyst in hedge fund, your task is to analyze the activities of hedge funds in q3 2025 by compare it with q2 2025. The dataset is downloaded to `/root/2025-q2` and `/root/2025-q3` folders respectively.
Questions to be answered in this task:
1. In Q3, what's the AUM of Renaissance Technologies fou... | 21 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/sec-financial-report.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 355,962 | 2,282 | 8,108 |
skills-gpt-5.5-2026-04-23-oh-exam-block-sequencing | null | null | null | null | null | You are helping a university registrar schedule final exams. The exams are already been grouped into blocks. Your job is to assign the blocks to available slots so that students have as few tightly packed exams as possible.
The task uses a prepared exam scheduling instance, which has 24 exam blocks and 24 time slots.... | exam-block-sequencing | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/exam-block-sequencing","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/exam-block-seq... | You are helping a university registrar schedule final exams. The exams are already been grouped into blocks. Your job is to assign the blocks to available slots so that students have as few tightly packed exams as possible.
The task uses a prepared exam scheduling instance, which has 24 exam blocks and 24 time slots.... | 36 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/exam-block-sequencing.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 1,351,445 | 9,940 | 28,461 |
skills-gpt-5.5-2026-04-23-oh-energy-market-pricing | null | null | null | null | null | You’re a market analyst at a regional transmission organization. The system managers have noticed that there was a peculiar price increase at multiple buses in yesterday’s day-ahead market. They suspect that one transmission constraint is causing the congestion problem and would like you to carry out a precise “what if... | energy-market-pricing | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/energy-market-pricing","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/energy-market... | You’re a market analyst at a regional transmission organization. The system managers have noticed that there was a peculiar price increase at multiple buses in yesterday’s day-ahead market. They suspect that one transmission constraint is causing the congestion problem and would like you to carry out a precise “what if... | 15 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/energy-market-pricing.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 236,465 | 4,357 | 10,511 |
skills-gpt-5.5-2026-04-23-oh-financial-modeling-qa | null | null | null | null | null | Please analysis large-scale data file and give me answer for a financial data analysis question.
The data file is located at /root/data.xlsx.
The background of the question is located at /root/background.pdf.
The question is: If the odd numbered games are played by Player 1, and the even numbered games are played by... | financial-modeling-qa | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/financial-modeling-qa","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/financial-mod... | Please analysis large-scale data file and give me answer for a financial data analysis question.
The data file is located at /root/data.xlsx.
The background of the question is located at /root/background.pdf.
The question is: If the odd numbered games are played by Player 1, and the even numbered games are played by... | 24 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/financial-modeling-qa.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 632,763 | 3,929 | 10,533 |
skills-gpt-5.5-2026-04-23-oh-energy-ac-optimal-power-flow | null | null | null | null | null | As an operator-planner in an Independent System Operator, you create a base case for tomorrow's peak hour each morning in order to establish the voltage profile and verify that the day-ahead market schedule is AC-feasible. In order to do this, you should use the model described in `math-model.md` to find a least-cost, ... | energy-ac-optimal-power-flow | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/energy-ac-optimal-power-flow","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/energy... | As an operator-planner in an Independent System Operator, you create a base case for tomorrow's peak hour each morning in order to establish the voltage profile and verify that the day-ahead market schedule is AC-feasible. In order to do this, you should use the model described in `math-model.md` to find a least-cost, ... | 21 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/energy-ac-optimal-power-flow.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 387,596 | 2,678 | 10,094 |
skills-gpt-5.5-2026-04-23-oh-crystallographic-wyckoff-position-analysis | null | null | null | null | null | You are a materials science researcher. You want to analyze the crystal structure of a few samples. You are given a few CIF files generated by SHELX through X-ray Diffraction.
To process the given CIF file data in batches, you are required to build a function in python to analyze Wyckoff position multiplicities and ca... | crystallographic-wyckoff-position-analysis | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/crystallographic-wyckoff-position-analysis","passed":false,"reward":0.55,"task_image":"openhands-ski... | You are a materials science researcher. You want to analyze the crystal structure of a few samples. You are given a few CIF files generated by SHELX through X-ray Diffraction.
To process the given CIF file data in batches, you are required to build a function in python to analyze Wyckoff position multiplicities and ca... | 16 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/crystallographic-wyckoff-position-analysis.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 232,887 | 3,301 | 6,607 |
skills-gpt-5.5-2026-04-23-oh-flink-query | null | null | null | null | null | In `/app/workspace/` I provided a skeleton code of a flink job and a dataset.
The provided dataset files are under `/app/workspace/data/`. It is a subset of the Google cluster-usage traces dataset. Read `/app/workspace/data/format.pdf` and `/app/workspace/data/ClusterData2011_2.md` to understand the data schema. The d... | flink-query | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/flink-query","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/flink-query:9f00f0626f0... | In `/app/workspace/` I provided a skeleton code of a flink job and a dataset.
The provided dataset files are under `/app/workspace/data/`. It is a subset of the Google cluster-usage traces dataset. Read `/app/workspace/data/format.pdf` and `/app/workspace/data/ClusterData2011_2.md` to understand the data schema. The d... | 37 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/flink-query.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 934,613 | 8,565 | 17,552 |
skills-gpt-5.5-2026-04-23-oh-llm-prefix-cache-replay | null | null | null | null | null | Replay the LLM inference request trace at `/root/trace.jsonl` through a block-level KV prefix cache and report hit statistics. Each line of `trace.jsonl` is one request with fields `timestamp`, `input_length`, `output_length`, and `hash_ids` — the prompt's block hashes in order.
Cache parameters are in `/root/config.j... | llm-prefix-cache-replay | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/llm-prefix-cache-replay","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/llm-prefix-c... | Replay the LLM inference request trace at `/root/trace.jsonl` through a block-level KV prefix cache and report hit statistics. Each line of `trace.jsonl` is one request with fields `timestamp`, `input_length`, `output_length`, and `hash_ids` — the prompt's block hashes in order.
Cache parameters are in `/root/config.j... | 6 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/llm-prefix-cache-replay.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 66,463 | 816 | 2,779 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-python-scala-translation | null | null | null | null | null | # Python to Scala Code Translation
'/root/Tokenizer.py' is a python code for data preparation, and you need to translate this code into Scala for processing massive data in distributed systems. You will need to make sure your Scala code follows the best practices and save your file in `/root/Tokenizer.scala`. Your Sca... | python-scala-translation | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/python-scala-translation/critic_attempt_1/run_fb305a2f... | # Python to Scala Code Translation
'/root/Tokenizer.py' is a python code for data preparation, and you need to translate this code into Scala for processing massive data in distributed systems. You will need to make sure your Scala code follows the best practices and save your file in `/root/Tokenizer.scala`. Your Sca... | 43 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/python-scala-translation.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 1,395,955 | 48,654 | 62,750 |
skills-gpt-5.5-2026-04-23-oh-pptx-reference-formatting | null | null | null | null | null | Help me detect all dangling paper titles in the slides `/root/Awesome-Agent-Papers.pptx`, and do the following
- change the font type to Arial, font size to 16, font color to #989596, and disable bold if any
- adjust the box width, so that each title in displayed in one line
- put the title at the bottom center of each... | pptx-reference-formatting | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/pptx-reference-formatting","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/pptx-refer... | Help me detect all dangling paper titles in the slides `/root/Awesome-Agent-Papers.pptx`, and do the following
- change the font type to Arial, font size to 16, font color to #989596, and disable bold if any
- adjust the box width, so that each title in displayed in one line
- put the title at the bottom center of each... | 20 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/pptx-reference-formatting.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 425,530 | 2,955 | 9,109 |
skills-gpt-5.5-2026-04-23-oh-mario-coin-counting | null | null | null | null | null | In this task, you are given a clip of screen recording of a player playing the game Super Mario. Your goal is to analyze the key frames in this video and count how many coins / enemies / turtles ever showing up in each frame and write the results into a CSV file located in `/root/counting_results.csv`.
Step 1. The vid... | mario-coin-counting | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/mario-coin-counting","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/mario-coin-coun... | In this task, you are given a clip of screen recording of a player playing the game Super Mario. Your goal is to analyze the key frames in this video and count how many coins / enemies / turtles ever showing up in each frame and write the results into a CSV file located in `/root/counting_results.csv`.
Step 1. The vid... | 23 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/mario-coin-counting.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 288,166 | 2,557 | 6,557 |
skills-gpt-5.5-2026-04-23-oh-reserves-at-risk-calc | null | null | null | null | null | STEP 1: download global commodity excel database from https://www.imf.org/en/research/commodity-prices, extract gold price (US$ per troy ounce) and populate the data into "Gold price" sheet. Calculate the monthly log, 3-month volatility and 12-month volatility in the relevant column (column C D E). Go to sheet "Answer"... | reserves-at-risk-calc | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/reserves-at-risk-calc","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/reserves-at-r... | STEP 1: download global commodity excel database from https://www.imf.org/en/research/commodity-prices, extract gold price (US$ per troy ounce) and populate the data into "Gold price" sheet. Calculate the monthly log, 3-month volatility and 12-month volatility in the relevant column (column C D E). Go to sheet "Answer"... | 32 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/reserves-at-risk-calc.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 893,585 | 10,907 | 19,712 |
skills-gemini-gemini-3.5-flash-oh-energy-market-pricing | null | null | null | null | null | You’re a market analyst at a regional transmission organization. The system managers have noticed that there was a peculiar price increase at multiple buses in yesterday’s day-ahead market. They suspect that one transmission constraint is causing the congestion problem and would like you to carry out a precise “what if... | energy-market-pricing | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/energy-market-pricing","passed":false,"reward":0.0,"task_image":"openhands-skillsbench... | You’re a market analyst at a regional transmission organization. The system managers have noticed that there was a peculiar price increase at multiple buses in yesterday’s day-ahead market. They suspect that one transmission constraint is causing the congestion problem and would like you to carry out a precise “what if... | 50 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/energy-market-pricing.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 1,654,855 | 31,229 | 47,585 |
skills-gemini-gemini-3.5-flash-oh-jpg-ocr-stat | null | null | null | null | null | ## Task description
In `/app/workspace/dataset/img`, I provide a set of scanned receipt images. Each receipt image contains text such as date, product name, unit price, total amount cost, etc. The text mainly consists of digits and English characters.
Read all image files under the given path, extract their data and ... | jpg-ocr-stat | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/jpg-ocr-stat","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/jpg-ocr-... | ## Task description
In `/app/workspace/dataset/img`, I provide a set of scanned receipt images. Each receipt image contains text such as date, product name, unit price, total amount cost, etc. The text mainly consists of digits and English characters.
Read all image files under the given path, extract their data and ... | 100 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/jpg-ocr-stat.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 5,073,894 | 39,920 | 58,370 |
skills-gpt-5.5-2026-04-23-oh-manufacturing-equipment-maintenance | null | null | null | null | null | Automation engineers in manufacturing rely on the instructions and guidance provided by the reflow machine handbook to maintain reflow machines and solve error issues. In order to complete the root cause analysis and do necessary calculations, automation engineers also need to check MES system logs, thermocouple sensor... | manufacturing-equipment-maintenance | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/manufacturing-equipment-maintenance","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/... | Automation engineers in manufacturing rely on the instructions and guidance provided by the reflow machine handbook to maintain reflow machines and solve error issues. In order to complete the root cause analysis and do necessary calculations, automation engineers also need to check MES system logs, thermocouple sensor... | 35 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/manufacturing-equipment-maintenance.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 1,151,339 | 8,720 | 19,319 |
skills-gpt-5.5-2026-04-23-oh-lake-warming-attribution | null | null | null | null | null | My data is in `/root/data/`, which includes:
1. `water_temperature.csv`: Lake surface temperature (0-5m)
2. `climate.csv`: Climate variables
3. `land_cover.csv`: Land cover data
4. `hydrology.csv`: Hydrology data
First, I want to do a trend analysis to determine whether there is a long-term warming trend for the water... | lake-warming-attribution | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/lake-warming-attribution","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/lake-warmin... | My data is in `/root/data/`, which includes:
1. `water_temperature.csv`: Lake surface temperature (0-5m)
2. `climate.csv`: Climate variables
3. `land_cover.csv`: Land cover data
4. `hydrology.csv`: Hydrology data
First, I want to do a trend analysis to determine whether there is a long-term warming trend for the water... | 15 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/lake-warming-attribution.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 214,136 | 2,855 | 7,250 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-crystallographic-wyckoff-position-analysis | null | null | null | null | null | You are a materials science researcher. You want to analyze the crystal structure of a few samples. You are given a few CIF files generated by SHELX through X-ray Diffraction.
To process the given CIF file data in batches, you are required to build a function in python to analyze Wyckoff position multiplicities and ca... | crystallographic-wyckoff-position-analysis | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/crystallographic-wyckoff-position-analysis/critic_atte... | You are a materials science researcher. You want to analyze the crystal structure of a few samples. You are given a few CIF files generated by SHELX through X-ray Diffraction.
To process the given CIF file data in batches, you are required to build a function in python to analyze Wyckoff position multiplicities and ca... | 29 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/crystallographic-wyckoff-position-analysis.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 731,353 | 40,892 | 50,295 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-sec-financial-report | null | null | null | null | null | You are a financial analyst in hedge fund, your task is to analyze the activities of hedge funds in q3 2025 by compare it with q2 2025. The dataset is downloaded to `/root/2025-q2` and `/root/2025-q3` folders respectively.
Questions to be answered in this task:
1. In Q3, what's the AUM of Renaissance Technologies fou... | sec-financial-report | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/sec-financial-report/critic_attempt_1/run_1b19bd8a4eae... | You are a financial analyst in hedge fund, your task is to analyze the activities of hedge funds in q3 2025 by compare it with q2 2025. The dataset is downloaded to `/root/2025-q2` and `/root/2025-q3` folders respectively.
Questions to be answered in this task:
1. In Q3, what's the AUM of Renaissance Technologies fou... | 23 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/sec-financial-report.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 346,566 | 2,607 | 6,714 |
skills-gpt-5.5-2026-04-23-oh-weighted-gdp-calc | null | null | null | null | null | You need to calculate the weighted mean of net exports as a percentage of GDP for the GCC in the file `gdp.xlsx`. The workbook contains two sheets: "Task" (where you will enter formulas) and "Data" (source data). Please work only within these existing sheets.
Step 1: The data you need for this problem are located in s... | weighted-gdp-calc | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/weighted-gdp-calc","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/weighted-gdp-calc:... | You need to calculate the weighted mean of net exports as a percentage of GDP for the GCC in the file `gdp.xlsx`. The workbook contains two sheets: "Task" (where you will enter formulas) and "Data" (source data). Please work only within these existing sheets.
Step 1: The data you need for this problem are located in s... | 24 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/weighted-gdp-calc.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 590,851 | 3,150 | 9,281 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-pddl-tpp-planning | null | null | null | null | null | Solve travelling purchase problem (TPP) tasks using PDDL (Planning Domain Definition Language).
Each task has two input files: a PDDL domain file and a PDDL problem file. As a planning agent, you may need both of them. The domain and problem file paths for each task are specified the "domain" key and "problem" key in ... | pddl-tpp-planning | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/pddl-tpp-planning/critic_attempt_1/run_bc06c2b6b43c49a... | Solve travelling purchase problem (TPP) tasks using PDDL (Planning Domain Definition Language).
Each task has two input files: a PDDL domain file and a PDDL problem file. As a planning agent, you may need both of them. The domain and problem file paths for each task are specified the "domain" key and "problem" key in ... | 21 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/pddl-tpp-planning.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 242,647 | 1,021 | 4,364 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-hvac-control | null | null | null | null | null | You need to implement a temperature controller to maintain a temperature of 22.0C. And the targets are: steady-state error <0.5C, settling time <120s, overshoot <10% , control duration >=150s, max temperature <30C. Also consider the constraints: Initial temperature ~18.0C (+/-2C sensor noise), heater power 0-100%. Simu... | hvac-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/hvac-control/critic_attempt_1/run_801d54b39d6849dea144... | You need to implement a temperature controller to maintain a temperature of 22.0C. And the targets are: steady-state error <0.5C, settling time <120s, overshoot <10% , control duration >=150s, max temperature <30C. Also consider the constraints: Initial temperature ~18.0C (+/-2C sensor noise), heater power 0-100%. Simu... | 15 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/hvac-control.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 190,155 | 14,254 | 20,196 |
skills-gemini-gemini-3.5-flash-oh-grid-dispatch-operator | null | null | null | null | null | You are a Grid Dispatch Operator. Given a power network snapshot, you want to decide the generators' dispatches. The dispatches should be economically efficient while satisfying the energy market requirement:
1. DC power balance at each bus
2. Generator and transmission limits
3. Spinning reserve requirements with stan... | grid-dispatch-operator | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/grid-dispatch-operator","passed":true,"reward":1.0,"task_image":"openhands-skillsbench... | You are a Grid Dispatch Operator. Given a power network snapshot, you want to decide the generators' dispatches. The dispatches should be economically efficient while satisfying the energy market requirement:
1. DC power balance at each bus
2. Generator and transmission limits
3. Spinning reserve requirements with stan... | 55 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/grid-dispatch-operator.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 2,725,912 | 24,622 | 40,009 |
skills-gemini-gemini-3.5-flash-oh-llm-prefix-cache-replay | null | null | null | null | null | Replay the LLM inference request trace at `/root/trace.jsonl` through a block-level KV prefix cache and report hit statistics. Each line of `trace.jsonl` is one request with fields `timestamp`, `input_length`, `output_length`, and `hash_ids` — the prompt's block hashes in order.
Cache parameters are in `/root/config.j... | llm-prefix-cache-replay | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/llm-prefix-cache-replay","passed":true,"reward":1.0,"task_image":"openhands-skillsbenc... | Replay the LLM inference request trace at `/root/trace.jsonl` through a block-level KV prefix cache and report hit statistics. Each line of `trace.jsonl` is one request with fields `timestamp`, `input_length`, `output_length`, and `hash_ids` — the prompt's block hashes in order.
Cache parameters are in `/root/config.j... | 12 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/llm-prefix-cache-replay.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 237,020 | 16,126 | 20,012 |
skills-gpt-5.5-2026-04-23-oh-manufacturing-codebook-normalization | null | null | null | null | null | At manufacturing test centers, testing engineers often write recognized defect reasons quickly with typos, noise, abbreviations, Chinese-English characters mixtures, etc. These texts vary largely between different testing engineers. Testing engineers are given standard codebooks but they may not follow this standardiza... | manufacturing-codebook-normalization | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/manufacturing-codebook-normalization","passed":false,"reward":0.0,"task_image":"openhands-skillsbenc... | At manufacturing test centers, testing engineers often write recognized defect reasons quickly with typos, noise, abbreviations, Chinese-English characters mixtures, etc. These texts vary largely between different testing engineers. Testing engineers are given standard codebooks but they may not follow this standardiza... | 26 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/manufacturing-codebook-normalization.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 1,595,740 | 9,865 | 30,721 |
skills-gemini-gemini-3.5-flash-oh-exam-block-sequencing | null | null | null | null | null | You are helping a university registrar schedule final exams. The exams are already been grouped into blocks. Your job is to assign the blocks to available slots so that students have as few tightly packed exams as possible.
The task uses a prepared exam scheduling instance, which has 24 exam blocks and 24 time slots.... | exam-block-sequencing | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/exam-block-sequencing","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/... | You are helping a university registrar schedule final exams. The exams are already been grouped into blocks. Your job is to assign the blocks to available slots so that students have as few tightly packed exams as possible.
The task uses a prepared exam scheduling instance, which has 24 exam blocks and 24 time slots.... | 42 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/exam-block-sequencing.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 1,814,258 | 36,600 | 59,735 |
skills-gpt-5.5-2026-04-23-oh-pddl-tpp-planning | null | null | null | null | null | Solve travelling purchase problem (TPP) tasks using PDDL (Planning Domain Definition Language).
Each task has two input files: a PDDL domain file and a PDDL problem file. As a planning agent, you may need both of them. The domain and problem file paths for each task are specified the "domain" key and "problem" key in ... | pddl-tpp-planning | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/pddl-tpp-planning","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/pddl-tpp-planning:... | Solve travelling purchase problem (TPP) tasks using PDDL (Planning Domain Definition Language).
Each task has two input files: a PDDL domain file and a PDDL problem file. As a planning agent, you may need both of them. The domain and problem file paths for each task are specified the "domain" key and "problem" key in ... | 11 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/pddl-tpp-planning.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 73,144 | 1,034 | 2,282 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-mario-coin-counting | null | null | null | null | null | In this task, you are given a clip of screen recording of a player playing the game Super Mario. Your goal is to analyze the key frames in this video and count how many coins / enemies / turtles ever showing up in each frame and write the results into a CSV file located in `/root/counting_results.csv`.
Step 1. The vid... | mario-coin-counting | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/mario-coin-counting/critic_attempt_1/run_f766263632a44... | In this task, you are given a clip of screen recording of a player playing the game Super Mario. Your goal is to analyze the key frames in this video and count how many coins / enemies / turtles ever showing up in each frame and write the results into a CSV file located in `/root/counting_results.csv`.
Step 1. The vid... | 77 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/mario-coin-counting.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 1,044,484 | 20,824 | 35,353 |
skills-gemini-gemini-3.5-flash-oh-lab-unit-harmonization | null | null | null | null | null | You are working on the clinical lab data. The data is actually consist of multiple source coming from different heathcare systems. That means they might using different units for the same blood test. I need your help conducting unit harmonization! Be careful about the in data format and inconsistent units used.
The in... | lab-unit-harmonization | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/lab-unit-harmonization","passed":false,"reward":0.667,"task_image":"openhands-skillsbe... | You are working on the clinical lab data. The data is actually consist of multiple source coming from different heathcare systems. That means they might using different units for the same blood test. I need your help conducting unit harmonization! Be careful about the in data format and inconsistent units used.
The in... | 52 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/lab-unit-harmonization.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 2,395,271 | 40,000 | 53,394 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-r2r-mpc-control | null | null | null | null | null | The task is designed for a 6-section Roll-to-Roll manufacturing line. You need to implement an MPC controller to control and make web tensions stable during section 3 roller changing from 20N to 44N at t=0.5. The simulator environment is r2r_simulator.py. Do not modify r2r_simulator.py. Controller must work with the or... | r2r-mpc-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/r2r-mpc-control/critic_attempt_1/run_3219e8b46edb4eaf9... | The task is designed for a 6-section Roll-to-Roll manufacturing line. You need to implement an MPC controller to control and make web tensions stable during section 3 roller changing from 20N to 44N at t=0.5. The simulator environment is r2r_simulator.py. Do not modify r2r_simulator.py. Controller must work with the or... | 46 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/r2r-mpc-control.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 920,352 | 46,948 | 73,203 |
skills-gpt-5.5-2026-04-23-oh-react-performance-debugging | null | null | null | null | null | There's an e-commerce website implemented in Next.js that is extremely slow and users keep complaining about it. Some scenarios that users have been complaining about:
- Adding items to cart is slow
- Compare page is extremely slow on load time
- api responses are slower than expected
Your job is to use best practice... | react-performance-debugging | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/react-performance-debugging","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/react-p... | There's an e-commerce website implemented in Next.js that is extremely slow and users keep complaining about it. Some scenarios that users have been complaining about:
- Adding items to cart is slow
- Compare page is extremely slow on load time
- api responses are slower than expected
Your job is to use best practice... | 50 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/react-performance-debugging.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 2,003,943 | 5,833 | 33,195 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-flood-risk-analysis | null | null | null | null | null | Find stations experienced flooding during April 1-7, 2025. USGS stations data is in `/root/data/michigan_stations.txt` which contains streamflow records. You should output a CSV file to `/root/output/flood_results.csv`. The 'flood_results.csv' should have two columns: `station_id` (the USGS station ID) and `flood_days... | flood-risk-analysis | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/flood-risk-analysis/critic_attempt_1/run_ad1e6884b9d24... | Find stations experienced flooding during April 1-7, 2025. USGS stations data is in `/root/data/michigan_stations.txt` which contains streamflow records. You should output a CSV file to `/root/output/flood_results.csv`. The 'flood_results.csv' should have two columns: `station_id` (the USGS station ID) and `flood_days... | 26 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/flood-risk-analysis.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 335,096 | 3,612 | 9,126 |
skills-gpt-5.5-2026-04-23-oh-python-scala-translation | null | null | null | null | null | # Python to Scala Code Translation
'/root/Tokenizer.py' is a python code for data preparation, and you need to translate this code into Scala for processing massive data in distributed systems. You will need to make sure your Scala code follows the best practices and save your file in `/root/Tokenizer.scala`. Your Sca... | python-scala-translation | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/python-scala-translation","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/python-scal... | # Python to Scala Code Translation
'/root/Tokenizer.py' is a python code for data preparation, and you need to translate this code into Scala for processing massive data in distributed systems. You will need to make sure your Scala code follows the best practices and save your file in `/root/Tokenizer.scala`. Your Sca... | 30 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/python-scala-translation.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 807,544 | 9,509 | 18,792 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-manufacturing-equipment-maintenance | null | null | null | null | null | Automation engineers in manufacturing rely on the instructions and guidance provided by the reflow machine handbook to maintain reflow machines and solve error issues. In order to complete the root cause analysis and do necessary calculations, automation engineers also need to check MES system logs, thermocouple sensor... | manufacturing-equipment-maintenance | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/manufacturing-equipment-maintenance/critic_attempt_1/r... | Automation engineers in manufacturing rely on the instructions and guidance provided by the reflow machine handbook to maintain reflow machines and solve error issues. In order to complete the root cause analysis and do necessary calculations, automation engineers also need to check MES system logs, thermocouple sensor... | 40 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/manufacturing-equipment-maintenance.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 1,748,538 | 40,171 | 53,631 |
skills-gemini-gemini-3.5-flash-oh-mario-coin-counting | null | null | null | null | null | In this task, you are given a clip of screen recording of a player playing the game Super Mario. Your goal is to analyze the key frames in this video and count how many coins / enemies / turtles ever showing up in each frame and write the results into a CSV file located in `/root/counting_results.csv`.
Step 1. The vid... | mario-coin-counting | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/mario-coin-counting","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/m... | In this task, you are given a clip of screen recording of a player playing the game Super Mario. Your goal is to analyze the key frames in this video and count how many coins / enemies / turtles ever showing up in each frame and write the results into a CSV file located in `/root/counting_results.csv`.
Step 1. The vid... | 29 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/mario-coin-counting.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 581,081 | 14,089 | 22,165 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-llm-prefix-cache-replay | null | null | null | null | null | Replay the LLM inference request trace at `/root/trace.jsonl` through a block-level KV prefix cache and report hit statistics. Each line of `trace.jsonl` is one request with fields `timestamp`, `input_length`, `output_length`, and `hash_ids` — the prompt's block hashes in order.
Cache parameters are in `/root/config.j... | llm-prefix-cache-replay | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/llm-prefix-cache-replay/critic_attempt_1/run_6f4746763... | Replay the LLM inference request trace at `/root/trace.jsonl` through a block-level KV prefix cache and report hit statistics. Each line of `trace.jsonl` is one request with fields `timestamp`, `input_length`, `output_length`, and `hash_ids` — the prompt's block hashes in order.
Cache parameters are in `/root/config.j... | 10 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/llm-prefix-cache-replay.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 129,016 | 7,123 | 13,300 |
skills-gemini-gemini-3.5-flash-oh-latex-formula-extraction | null | null | null | null | null | You are helping to extract all latex formulas in its own line from a research paper.
The paper is (`latex_paper.pdf`).
Your task is to:
1. Identify all latex formulas in its own line in each page (note that not every page has formulas in its own line)
2. Extract them as raw latex text string using `$$ ... $$` method... | latex-formula-extraction | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/latex-formula-extraction","passed":true,"reward":1.0,"task_image":"openhands-skillsben... | You are helping to extract all latex formulas in its own line from a research paper.
The paper is (`latex_paper.pdf`).
Your task is to:
1. Identify all latex formulas in its own line in each page (note that not every page has formulas in its own line)
2. Extract them as raw latex text string using `$$ ... $$` method... | 50 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/latex-formula-extraction.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 2,267,436 | 27,331 | 34,845 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-jax-computing-basics | null | null | null | null | null | Given a set of tasks in "problem.json", complete all programming using JAX language correctly. For each task, read the input data, write JAX code to do some computation, then produce an output file and save it.
1. For each task, load the input data specified by the "input" key in "problem.json".
2. Perform the computa... | jax-computing-basics | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/jax-computing-basics/critic_attempt_1/run_daac18a5f151... | Given a set of tasks in "problem.json", complete all programming using JAX language correctly. For each task, read the input data, write JAX code to do some computation, then produce an output file and save it.
1. For each task, load the input data specified by the "input" key in "problem.json".
2. Perform the computa... | 15 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/jax-computing-basics.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 158,573 | 3,271 | 6,557 |
skills-gpt-5.5-2026-04-23-oh-lab-unit-harmonization | null | null | null | null | null | You are working on the clinical lab data. The data is actually consist of multiple source coming from different heathcare systems. That means they might using different units for the same blood test. I need your help conducting unit harmonization! Be careful about the in data format and inconsistent units used.
The in... | lab-unit-harmonization | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/lab-unit-harmonization","passed":false,"reward":0.833,"task_image":"openhands-skillsbench/lab-unit-h... | You are working on the clinical lab data. The data is actually consist of multiple source coming from different heathcare systems. That means they might using different units for the same blood test. I need your help conducting unit harmonization! Be careful about the in data format and inconsistent units used.
The in... | 13 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/lab-unit-harmonization.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 299,411 | 9,015 | 16,105 |
skills-gpt-5.5-2026-04-23-oh-radar-vital-signs | null | null | null | null | null | Task:
Estimate heart rate and breathing rate from 15 24 GHz continuous-wave radar recordings. The data was from a dataset where the radar is hanged 40cm in front of a subject who is lied on a tilt table.
Input:
The recordings are in `/root/recordings/`, named `rec_001` through `rec_015`. Each recording folder comes as... | radar-vital-signs | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/radar-vital-signs","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/radar-vital-signs:... | Task:
Estimate heart rate and breathing rate from 15 24 GHz continuous-wave radar recordings. The data was from a dataset where the radar is hanged 40cm in front of a subject who is lied on a tilt table.
Input:
The recordings are in `/root/recordings/`, named `rec_001` through `rec_015`. Each recording folder comes as... | 22 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/radar-vital-signs.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 959,438 | 13,447 | 25,438 |
skills-gemini-gemini-3.5-flash-oh-radar-vital-signs | null | null | null | null | null | Task:
Estimate heart rate and breathing rate from 15 24 GHz continuous-wave radar recordings. The data was from a dataset where the radar is hanged 40cm in front of a subject who is lied on a tilt table.
Input:
The recordings are in `/root/recordings/`, named `rec_001` through `rec_015`. Each recording folder comes as... | radar-vital-signs | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/radar-vital-signs","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/rada... | Task:
Estimate heart rate and breathing rate from 15 24 GHz continuous-wave radar recordings. The data was from a dataset where the radar is hanged 40cm in front of a subject who is lied on a tilt table.
Input:
The recordings are in `/root/recordings/`, named `rec_001` through `rec_015`. Each recording folder comes as... | 35 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/radar-vital-signs.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 994,974 | 15,133 | 25,646 |
skills-gpt-5.5-2026-04-23-oh-flood-risk-analysis | null | null | null | null | null | Find stations experienced flooding during April 1-7, 2025. USGS stations data is in `/root/data/michigan_stations.txt` which contains streamflow records. You should output a CSV file to `/root/output/flood_results.csv`. The 'flood_results.csv' should have two columns: `station_id` (the USGS station ID) and `flood_days... | flood-risk-analysis | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/flood-risk-analysis","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/flood-risk-anal... | Find stations experienced flooding during April 1-7, 2025. USGS stations data is in `/root/data/michigan_stations.txt` which contains streamflow records. You should output a CSV file to `/root/output/flood_results.csv`. The 'flood_results.csv' should have two columns: `station_id` (the USGS station ID) and `flood_days... | 15 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/flood-risk-analysis.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 208,346 | 2,168 | 5,913 |
skills-gpt-5.5-2026-04-23-oh-grid-dispatch-operator | null | null | null | null | null | You are a Grid Dispatch Operator. Given a power network snapshot, you want to decide the generators' dispatches. The dispatches should be economically efficient while satisfying the energy market requirement:
1. DC power balance at each bus
2. Generator and transmission limits
3. Spinning reserve requirements with stan... | grid-dispatch-operator | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/grid-dispatch-operator","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/grid-dispatc... | You are a Grid Dispatch Operator. Given a power network snapshot, you want to decide the generators' dispatches. The dispatches should be economically efficient while satisfying the energy market requirement:
1. DC power balance at each bus
2. Generator and transmission limits
3. Spinning reserve requirements with stan... | 16 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/grid-dispatch-operator.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 236,487 | 4,286 | 12,144 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-jpg-ocr-stat | null | null | null | null | null | ## Task description
In `/app/workspace/dataset/img`, I provide a set of scanned receipt images. Each receipt image contains text such as date, product name, unit price, total amount cost, etc. The text mainly consists of digits and English characters.
Read all image files under the given path, extract their data and ... | jpg-ocr-stat | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/jpg-ocr-stat/critic_attempt_1/run_c71502b8a3f14e9eaa40... | ## Task description
In `/app/workspace/dataset/img`, I provide a set of scanned receipt images. Each receipt image contains text such as date, product name, unit price, total amount cost, etc. The text mainly consists of digits and English characters.
Read all image files under the given path, extract their data and ... | 92 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/jpg-ocr-stat.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 4,636,152 | 243,788 | 284,151 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-energy-ac-optimal-power-flow | null | null | null | null | null | As an operator-planner in an Independent System Operator, you create a base case for tomorrow's peak hour each morning in order to establish the voltage profile and verify that the day-ahead market schedule is AC-feasible. In order to do this, you should use the model described in `math-model.md` to find a least-cost, ... | energy-ac-optimal-power-flow | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/energy-ac-optimal-power-flow/critic_attempt_1/run_f12b... | As an operator-planner in an Independent System Operator, you create a base case for tomorrow's peak hour each morning in order to establish the voltage profile and verify that the day-ahead market schedule is AC-feasible. In order to do this, you should use the model described in `math-model.md` to find a least-cost, ... | 86 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/energy-ac-optimal-power-flow.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 6,179,159 | 141,544 | 204,023 |
skills-gemini-gemini-3.5-flash-oh-shock-analysis-supply | null | null | null | null | null | This task is to estimate an investment spending shock to a small open economy (Georgia) using Cobb-Douglas production function to estimate the potential GDP. The investment will last 8 years beginning from 2026, which worths 6.5 billion USD. You are going to collect the relevant data from multiple online sources, popul... | shock-analysis-supply | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/shock-analysis-supply","passed":false,"reward":0.0,"task_image":"openhands-skillsbench... | This task is to estimate an investment spending shock to a small open economy (Georgia) using Cobb-Douglas production function to estimate the potential GDP. The investment will last 8 years beginning from 2026, which worths 6.5 billion USD. You are going to collect the relevant data from multiple online sources, popul... | 100 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/shock-analysis-supply.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 5,503,475 | 30,137 | 50,139 |
skills-gpt-5.5-2026-04-23-oh-adaptive-cruise-control | null | null | null | null | null | You need to implement an Adaptive Cruise Control (ACC) simulation that maintains the set speed (30m/s) when no vehicles are detected ahead, and automatically adjusts speed to maintain a safe following distance when a vehicle is detected ahead. The targets are: speed rise time <10s, speed overshoot <5%, speed steady-sta... | adaptive-cruise-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/adaptive-cruise-control","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/adaptive-cru... | You need to implement an Adaptive Cruise Control (ACC) simulation that maintains the set speed (30m/s) when no vehicles are detected ahead, and automatically adjusts speed to maintain a safe following distance when a vehicle is detected ahead. The targets are: speed rise time <10s, speed overshoot <5%, speed steady-sta... | 52 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/adaptive-cruise-control.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 2,177,381 | 8,205 | 28,843 |
skills-gpt-5.5-2026-04-23-oh-edit-pdf | null | null | null | null | null | You are given a `PDF` file at `/root/input/input.pdf` and a text file at `/root/input/input.txt` containing natural language instructions and updated information.
Your task is to update this PDF using the latest information provided in the text file and follow all the instructions in it.
Use the text file as the sourc... | edit-pdf | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/edit-pdf","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/edit-pdf:b17b2a1a8c7ff46d",... | You are given a `PDF` file at `/root/input/input.pdf` and a text file at `/root/input/input.txt` containing natural language instructions and updated information.
Your task is to update this PDF using the latest information provided in the text file and follow all the instructions in it.
Use the text file as the sourc... | 22 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/edit-pdf.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 653,222 | 7,535 | 13,967 |
skills-gpt-5.5-2026-04-23-oh-r2r-mpc-control | null | null | null | null | null | The task is designed for a 6-section Roll-to-Roll manufacturing line. You need to implement an MPC controller to control and make web tensions stable during section 3 roller changing from 20N to 44N at t=0.5. The simulator environment is r2r_simulator.py. Do not modify r2r_simulator.py. Controller must work with the or... | r2r-mpc-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/r2r-mpc-control","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/r2r-mpc-control:7cb... | The task is designed for a 6-section Roll-to-Roll manufacturing line. You need to implement an MPC controller to control and make web tensions stable during section 3 roller changing from 20N to 44N at t=0.5. The simulator environment is r2r_simulator.py. Do not modify r2r_simulator.py. Controller must work with the or... | 32 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/r2r-mpc-control.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 766,634 | 7,235 | 17,709 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-radar-vital-signs | null | null | null | null | null | Task:
Estimate heart rate and breathing rate from 15 24 GHz continuous-wave radar recordings. The data was from a dataset where the radar is hanged 40cm in front of a subject who is lied on a tilt table.
Input:
The recordings are in `/root/recordings/`, named `rec_001` through `rec_015`. Each recording folder comes as... | radar-vital-signs | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/radar-vital-signs/critic_attempt_1/run_176ab94f2cbf422... | Task:
Estimate heart rate and breathing rate from 15 24 GHz continuous-wave radar recordings. The data was from a dataset where the radar is hanged 40cm in front of a subject who is lied on a tilt table.
Input:
The recordings are in `/root/recordings/`, named `rec_001` through `rec_015`. Each recording folder comes as... | 24 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/radar-vital-signs.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 384,012 | 10,146 | 20,306 |
skills-gemini-gemini-3.5-flash-oh-edit-pdf | null | null | null | null | null | You are given a `PDF` file at `/root/input/input.pdf` and a text file at `/root/input/input.txt` containing natural language instructions and updated information.
Your task is to update this PDF using the latest information provided in the text file and follow all the instructions in it.
Use the text file as the sourc... | edit-pdf | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/edit-pdf","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/edit-pdf:b17b... | You are given a `PDF` file at `/root/input/input.pdf` and a text file at `/root/input/input.txt` containing natural language instructions and updated information.
Your task is to update this PDF using the latest information provided in the text file and follow all the instructions in it.
Use the text file as the sourc... | 32 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/edit-pdf.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 1,375,306 | 29,953 | 40,367 |
skills-gemini-gemini-3.5-flash-oh-r2r-mpc-control | null | null | null | null | null | The task is designed for a 6-section Roll-to-Roll manufacturing line. You need to implement an MPC controller to control and make web tensions stable during section 3 roller changing from 20N to 44N at t=0.5. The simulator environment is r2r_simulator.py. Do not modify r2r_simulator.py. Controller must work with the or... | r2r-mpc-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/r2r-mpc-control","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/r2r-m... | The task is designed for a 6-section Roll-to-Roll manufacturing line. You need to implement an MPC controller to control and make web tensions stable during section 3 roller changing from 20N to 44N at t=0.5. The simulator environment is r2r_simulator.py. Do not modify r2r_simulator.py. Controller must work with the or... | 40 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/r2r-mpc-control.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 1,677,971 | 27,735 | 45,603 |
skills-gemini-gemini-3.5-flash-oh-weighted-gdp-calc | null | null | null | null | null | You need to calculate the weighted mean of net exports as a percentage of GDP for the GCC in the file `gdp.xlsx`. The workbook contains two sheets: "Task" (where you will enter formulas) and "Data" (source data). Please work only within these existing sheets.
Step 1: The data you need for this problem are located in s... | weighted-gdp-calc | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/weighted-gdp-calc","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/weig... | You need to calculate the weighted mean of net exports as a percentage of GDP for the GCC in the file `gdp.xlsx`. The workbook contains two sheets: "Task" (where you will enter formulas) and "Data" (source data). Please work only within these existing sheets.
Step 1: The data you need for this problem are located in s... | 49 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/weighted-gdp-calc.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 1,937,151 | 28,081 | 37,863 |
skills-gpt-5.5-2026-04-23-oh-hvac-control | null | null | null | null | null | You need to implement a temperature controller to maintain a temperature of 22.0C. And the targets are: steady-state error <0.5C, settling time <120s, overshoot <10% , control duration >=150s, max temperature <30C. Also consider the constraints: Initial temperature ~18.0C (+/-2C sensor noise), heater power 0-100%. Simu... | hvac-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/hvac-control","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/hvac-control:dcaf46043... | You need to implement a temperature controller to maintain a temperature of 22.0C. And the targets are: steady-state error <0.5C, settling time <120s, overshoot <10% , control duration >=150s, max temperature <30C. Also consider the constraints: Initial temperature ~18.0C (+/-2C sensor noise), heater power 0-100%. Simu... | 13 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/hvac-control.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gpt-5.5-2026-04-23 | null | null | 131,755 | 1,339 | 5,959 |
skills-gpt-5.5-2026-04-23-oh-jax-computing-basics | null | null | null | null | null | Given a set of tasks in "problem.json", complete all programming using JAX language correctly. For each task, read the input data, write JAX code to do some computation, then produce an output file and save it.
1. For each task, load the input data specified by the "input" key in "problem.json".
2. Perform the computa... | jax-computing-basics | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/jax-computing-basics","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/jax-computing-b... | Given a set of tasks in "problem.json", complete all programming using JAX language correctly. For each task, read the input data, write JAX code to do some computation, then produce an output file and save it.
1. For each task, load the input data specified by the "input" key in "problem.json".
2. Perform the computa... | 15 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/jax-computing-basics.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 144,083 | 797 | 5,030 |
skills-gemini-gemini-3.5-flash-oh-hvac-control | null | null | null | null | null | You need to implement a temperature controller to maintain a temperature of 22.0C. And the targets are: steady-state error <0.5C, settling time <120s, overshoot <10% , control duration >=150s, max temperature <30C. Also consider the constraints: Initial temperature ~18.0C (+/-2C sensor noise), heater power 0-100%. Simu... | hvac-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/hvac-control","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/hvac-cont... | You need to implement a temperature controller to maintain a temperature of 22.0C. And the targets are: steady-state error <0.5C, settling time <120s, overshoot <10% , control duration >=150s, max temperature <30C. Also consider the constraints: Initial temperature ~18.0C (+/-2C sensor noise), heater power 0-100%. Simu... | 29 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/hvac-control.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 657,320 | 13,977 | 22,168 |
skills-gemini-gemini-3.5-flash-oh-manufacturing-equipment-maintenance | null | null | null | null | null | Automation engineers in manufacturing rely on the instructions and guidance provided by the reflow machine handbook to maintain reflow machines and solve error issues. In order to complete the root cause analysis and do necessary calculations, automation engineers also need to check MES system logs, thermocouple sensor... | manufacturing-equipment-maintenance | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/manufacturing-equipment-maintenance","passed":true,"reward":1.0,"task_image":"openhand... | Automation engineers in manufacturing rely on the instructions and guidance provided by the reflow machine handbook to maintain reflow machines and solve error issues. In order to complete the root cause analysis and do necessary calculations, automation engineers also need to check MES system logs, thermocouple sensor... | 69 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/manufacturing-equipment-maintenance.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 3,498,262 | 42,406 | 59,407 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-financial-modeling-qa | null | null | null | null | null | Please analysis large-scale data file and give me answer for a financial data analysis question.
The data file is located at /root/data.xlsx.
The background of the question is located at /root/background.pdf.
The question is: If the odd numbered games are played by Player 1, and the even numbered games are played by... | financial-modeling-qa | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/financial-modeling-qa/critic_attempt_1/run_0356285b693... | Please analysis large-scale data file and give me answer for a financial data analysis question.
The data file is located at /root/data.xlsx.
The background of the question is located at /root/background.pdf.
The question is: If the odd numbered games are played by Player 1, and the even numbered games are played by... | 15 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/financial-modeling-qa.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 243,559 | 6,455 | 12,348 |
skills-gemini-gemini-3.5-flash-oh-jax-computing-basics | null | null | null | null | null | Given a set of tasks in "problem.json", complete all programming using JAX language correctly. For each task, read the input data, write JAX code to do some computation, then produce an output file and save it.
1. For each task, load the input data specified by the "input" key in "problem.json".
2. Perform the computa... | jax-computing-basics | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/jax-computing-basics","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/j... | Given a set of tasks in "problem.json", complete all programming using JAX language correctly. For each task, read the input data, write JAX code to do some computation, then produce an output file and save it.
1. For each task, load the input data specified by the "input" key in "problem.json".
2. Perform the computa... | 22 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/jax-computing-basics.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 296,601 | 6,436 | 11,465 |
skills-gemini-gemini-3.5-flash-oh-manufacturing-codebook-normalization | null | null | null | null | null | At manufacturing test centers, testing engineers often write recognized defect reasons quickly with typos, noise, abbreviations, Chinese-English characters mixtures, etc. These texts vary largely between different testing engineers. Testing engineers are given standard codebooks but they may not follow this standardiza... | manufacturing-codebook-normalization | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/manufacturing-codebook-normalization","passed":false,"reward":0.0,"task_image":"openha... | At manufacturing test centers, testing engineers often write recognized defect reasons quickly with typos, noise, abbreviations, Chinese-English characters mixtures, etc. These texts vary largely between different testing engineers. Testing engineers are given standard codebooks but they may not follow this standardiza... | 73 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/manufacturing-codebook-normalization.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 5,496,656 | 53,527 | 104,014 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-weighted-gdp-calc | null | null | null | null | null | You need to calculate the weighted mean of net exports as a percentage of GDP for the GCC in the file `gdp.xlsx`. The workbook contains two sheets: "Task" (where you will enter formulas) and "Data" (source data). Please work only within these existing sheets.
Step 1: The data you need for this problem are located in s... | weighted-gdp-calc | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/weighted-gdp-calc/critic_attempt_1/run_bb06e698eeab48b... | You need to calculate the weighted mean of net exports as a percentage of GDP for the GCC in the file `gdp.xlsx`. The workbook contains two sheets: "Task" (where you will enter formulas) and "Data" (source data). Please work only within these existing sheets.
Step 1: The data you need for this problem are located in s... | 26 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/weighted-gdp-calc.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 671,407 | 12,559 | 19,959 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-adaptive-cruise-control | null | null | null | null | null | You need to implement an Adaptive Cruise Control (ACC) simulation that maintains the set speed (30m/s) when no vehicles are detected ahead, and automatically adjusts speed to maintain a safe following distance when a vehicle is detected ahead. The targets are: speed rise time <10s, speed overshoot <5%, speed steady-sta... | adaptive-cruise-control | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/adaptive-cruise-control/critic_attempt_1/run_e6e59b8a2... | You need to implement an Adaptive Cruise Control (ACC) simulation that maintains the set speed (30m/s) when no vehicles are detected ahead, and automatically adjusts speed to maintain a safe following distance when a vehicle is detected ahead. The targets are: speed rise time <10s, speed overshoot <5%, speed steady-sta... | 75 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/adaptive-cruise-control.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 3,014,639 | 67,422 | 98,243 |
skills-gemini-gemini-3.5-flash-oh-tictoc-unnecessary-abort-detection | null | null | null | null | null | The famous TicToc concurrency-control mechanism can sometimes abort transactions that did not actually need to abort and could have been left in execution. Your task is to identify the policy of detecting every unnecessary abort.
You are given a trace from a TicToc workload to test if you can detect unnecessary aborts... | tictoc-unnecessary-abort-detection | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/tictoc-unnecessary-abort-detection","passed":false,"reward":0.85,"task_image":"openhan... | The famous TicToc concurrency-control mechanism can sometimes abort transactions that did not actually need to abort and could have been left in execution. Your task is to identify the policy of detecting every unnecessary abort.
You are given a trace from a TicToc workload to test if you can detect unnecessary aborts... | 31 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/tictoc-unnecessary-abort-detection.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 1,415,637 | 25,512 | 34,464 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-latex-formula-extraction | null | null | null | null | null | You are helping to extract all latex formulas in its own line from a research paper.
The paper is (`latex_paper.pdf`).
Your task is to:
1. Identify all latex formulas in its own line in each page (note that not every page has formulas in its own line)
2. Extract them as raw latex text string using `$$ ... $$` method... | latex-formula-extraction | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/latex-formula-extraction/critic_attempt_1/run_98ef520f... | You are helping to extract all latex formulas in its own line from a research paper.
The paper is (`latex_paper.pdf`).
Your task is to:
1. Identify all latex formulas in its own line in each page (note that not every page has formulas in its own line)
2. Extract them as raw latex text string using `$$ ... $$` method... | 28 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/latex-formula-extraction.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 541,458 | 16,058 | 21,342 |
skills-gemini-gemini-3.5-flash-oh-flink-query | null | null | null | null | null | In `/app/workspace/` I provided a skeleton code of a flink job and a dataset.
The provided dataset files are under `/app/workspace/data/`. It is a subset of the Google cluster-usage traces dataset. Read `/app/workspace/data/format.pdf` and `/app/workspace/data/ClusterData2011_2.md` to understand the data schema. The d... | flink-query | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/flink-query","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/flink-que... | In `/app/workspace/` I provided a skeleton code of a flink job and a dataset.
The provided dataset files are under `/app/workspace/data/`. It is a subset of the Google cluster-usage traces dataset. Read `/app/workspace/data/format.pdf` and `/app/workspace/data/ClusterData2011_2.md` to understand the data schema. The d... | 65 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/flink-query.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 4,385,607 | 45,723 | 61,671 |
skills-gemini-gemini-3.5-flash-oh-sec-financial-report | null | null | null | null | null | You are a financial analyst in hedge fund, your task is to analyze the activities of hedge funds in q3 2025 by compare it with q2 2025. The dataset is downloaded to `/root/2025-q2` and `/root/2025-q3` folders respectively.
Questions to be answered in this task:
1. In Q3, what's the AUM of Renaissance Technologies fou... | sec-financial-report | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/sec-financial-report","passed":false,"reward":0.0,"task_image":"openhands-skillsbench/... | You are a financial analyst in hedge fund, your task is to analyze the activities of hedge funds in q3 2025 by compare it with q2 2025. The dataset is downloaded to `/root/2025-q2` and `/root/2025-q3` folders respectively.
Questions to be answered in this task:
1. In Q3, what's the AUM of Renaissance Technologies fou... | 51 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/sec-financial-report.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 1,969,282 | 20,756 | 34,060 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-react-performance-debugging | null | null | null | null | null | There's an e-commerce website implemented in Next.js that is extremely slow and users keep complaining about it. Some scenarios that users have been complaining about:
- Adding items to cart is slow
- Compare page is extremely slow on load time
- api responses are slower than expected
Your job is to use best practice... | react-performance-debugging | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/react-performance-debugging/critic_attempt_1/run_9ae6a... | There's an e-commerce website implemented in Next.js that is extremely slow and users keep complaining about it. Some scenarios that users have been complaining about:
- Adding items to cart is slow
- Compare page is extremely slow on load time
- api responses are slower than expected
Your job is to use best practice... | 76 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/react-performance-debugging.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 2,483,344 | 41,393 | 65,451 |
skills-gemini-gemini-3.5-flash-oh-pddl-tpp-planning | null | null | null | null | null | Solve travelling purchase problem (TPP) tasks using PDDL (Planning Domain Definition Language).
Each task has two input files: a PDDL domain file and a PDDL problem file. As a planning agent, you may need both of them. The domain and problem file paths for each task are specified the "domain" key and "problem" key in ... | pddl-tpp-planning | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/pddl-tpp-planning","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/pddl... | Solve travelling purchase problem (TPP) tasks using PDDL (Planning Domain Definition Language).
Each task has two input files: a PDDL domain file and a PDDL problem file. As a planning agent, you may need both of them. The domain and problem file paths for each task are specified the "domain" key and "problem" key in ... | 26 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/pddl-tpp-planning.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 439,348 | 5,846 | 9,906 |
skills-gemini-gemini-3.5-flash-oh-invoice-fraud-detection | null | null | null | null | null | You need to analyze the following files to find any potential invoice fraud:
- `/root/invoices.pdf`: One invoice per page.
- `/root/vendors.xlsx`: Approved vendors with corresponding Vendor ID, Name, and authorized IBAN.
- `/root/purchase_orders.csv`: Valid purchase orders with corresponding PO Number, Amount, and Vend... | invoice-fraud-detection | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/invoice-fraud-detection","passed":true,"reward":1.0,"task_image":"openhands-skillsbenc... | You need to analyze the following files to find any potential invoice fraud:
- `/root/invoices.pdf`: One invoice per page.
- `/root/vendors.xlsx`: Approved vendors with corresponding Vendor ID, Name, and authorized IBAN.
- `/root/purchase_orders.csv`: Valid purchase orders with corresponding PO Number, Amount, and Vend... | 40 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/invoice-fraud-detection.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 1,037,921 | 18,031 | 28,701 |
skills-gemini-gemini-3.5-flash-oh-earthquake-plate-calculation | null | null | null | null | null | You are a geospatial analyst with expertise in plate tectonics and earthquake analysis. I want to find the earthquake that occurred furthest from the Pacific plate boundary within the Pacific plate itself. Use GeoPandas projections. Output the result to `/root/answer.json` as a JSON file with the following fields:
- `... | earthquake-plate-calculation | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/earthquake-plate-calculation","passed":false,"reward":0.0,"task_image":"openhands-skil... | You are a geospatial analyst with expertise in plate tectonics and earthquake analysis. I want to find the earthquake that occurred furthest from the Pacific plate boundary within the Pacific plate itself. Use GeoPandas projections. Output the result to `/root/answer.json` as a JSON file with the following fields:
- `... | 43 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/earthquake-plate-calculation.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 1,606,229 | 23,559 | 40,843 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-invoice-fraud-detection | null | null | null | null | null | You need to analyze the following files to find any potential invoice fraud:
- `/root/invoices.pdf`: One invoice per page.
- `/root/vendors.xlsx`: Approved vendors with corresponding Vendor ID, Name, and authorized IBAN.
- `/root/purchase_orders.csv`: Valid purchase orders with corresponding PO Number, Amount, and Vend... | invoice-fraud-detection | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/invoice-fraud-detection/critic_attempt_1/run_0a3f106a2... | You need to analyze the following files to find any potential invoice fraud:
- `/root/invoices.pdf`: One invoice per page.
- `/root/vendors.xlsx`: Approved vendors with corresponding Vendor ID, Name, and authorized IBAN.
- `/root/purchase_orders.csv`: Valid purchase orders with corresponding PO Number, Amount, and Vend... | 36 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/invoice-fraud-detection.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 1,109,587 | 14,170 | 22,208 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-earthquake-plate-calculation | null | null | null | null | null | You are a geospatial analyst with expertise in plate tectonics and earthquake analysis. I want to find the earthquake that occurred furthest from the Pacific plate boundary within the Pacific plate itself. Use GeoPandas projections. Output the result to `/root/answer.json` as a JSON file with the following fields:
- `... | earthquake-plate-calculation | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/earthquake-plate-calculation/critic_attempt_1/run_c9fe... | You are a geospatial analyst with expertise in plate tectonics and earthquake analysis. I want to find the earthquake that occurred furthest from the Pacific plate boundary within the Pacific plate itself. Use GeoPandas projections. Output the result to `/root/answer.json` as a JSON file with the following fields:
- `... | 14 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/earthquake-plate-calculation.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 158,499 | 1,926 | 4,767 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-ada-bathroom-plan-repair | null | null | null | null | null | # Task
You are given a standardized 2D toilet-room CAD plan. Extract the original room layout, identify the plan-view ADA (Americans with Disabilities Act)-derived accessibility violations, and produce a minimally invasive repaired CAD layout that satisfies the provided rules while preserving the original design where... | ada-bathroom-plan-repair | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/ada-bathroom-plan-repair/critic_attempt_1/run_08039795... | # Task
You are given a standardized 2D toilet-room CAD plan. Extract the original room layout, identify the plan-view ADA (Americans with Disabilities Act)-derived accessibility violations, and produce a minimally invasive repaired CAD layout that satisfies the provided rules while preserving the original design where... | 27 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/ada-bathroom-plan-repair.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 938,588 | 91,347 | 105,016 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-pptx-reference-formatting | null | null | null | null | null | Help me detect all dangling paper titles in the slides `/root/Awesome-Agent-Papers.pptx`, and do the following
- change the font type to Arial, font size to 16, font color to #989596, and disable bold if any
- adjust the box width, so that each title in displayed in one line
- put the title at the bottom center of each... | pptx-reference-formatting | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/pptx-reference-formatting/critic_attempt_1/run_b20dc3b... | Help me detect all dangling paper titles in the slides `/root/Awesome-Agent-Papers.pptx`, and do the following
- change the font type to Arial, font size to 16, font color to #989596, and disable bold if any
- adjust the box width, so that each title in displayed in one line
- put the title at the bottom center of each... | 17 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/pptx-reference-formatting.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 299,530 | 5,910 | 10,534 |
skills-gemini-gemini-3.5-flash-oh-reserves-at-risk-calc | null | null | null | null | null | STEP 1: download global commodity excel database from https://www.imf.org/en/research/commodity-prices, extract gold price (US$ per troy ounce) and populate the data into "Gold price" sheet. Calculate the monthly log, 3-month volatility and 12-month volatility in the relevant column (column C D E). Go to sheet "Answer"... | reserves-at-risk-calc | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/reserves-at-risk-calc","passed":false,"reward":0.0,"task_image":"openhands-skillsbench... | STEP 1: download global commodity excel database from https://www.imf.org/en/research/commodity-prices, extract gold price (US$ per troy ounce) and populate the data into "Gold price" sheet. Calculate the monthly log, 3-month volatility and 12-month volatility in the relevant column (column C D E). Go to sheet "Answer"... | 46 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/reserves-at-risk-calc.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | gemini/gemini-3.5-flash | null | null | 1,955,567 | 41,307 | 60,720 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-lab-unit-harmonization | null | null | null | null | null | You are working on the clinical lab data. The data is actually consist of multiple source coming from different heathcare systems. That means they might using different units for the same blood test. I need your help conducting unit harmonization! Be careful about the in data format and inconsistent units used.
The in... | lab-unit-harmonization | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/lab-unit-harmonization/critic_attempt_1/run_3263eab1c7... | You are working on the clinical lab data. The data is actually consist of multiple source coming from different heathcare systems. That means they might using different units for the same blood test. I need your help conducting unit harmonization! Be careful about the in data format and inconsistent units used.
The in... | 55 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/lab-unit-harmonization.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 2,220,167 | 84,775 | 122,916 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-fix-erlang-ssh-cve | null | null | null | null | null | Erlang/OTP SSH is the built-in SSH server component of the Erlang/OTP platform.
A critical vulnerability was discovered in the Erlang/OTP SSH server, allowing attackers to execute arbitrary system commands remotely without authentication by crafting specific SSH protocol messages. In `/app/workspace/otp_src_27.3.2` we... | fix-erlang-ssh-cve | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/fix-erlang-ssh-cve/critic_attempt_1/run_ac9d353abe2842... | Erlang/OTP SSH is the built-in SSH server component of the Erlang/OTP platform.
A critical vulnerability was discovered in the Erlang/OTP SSH server, allowing attackers to execute arbitrary system commands remotely without authentication by crafting specific SSH protocol messages. In `/app/workspace/otp_src_27.3.2` we... | 37 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/fix-erlang-ssh-cve.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 938,435 | 24,973 | 32,093 |
skills-gemini-gemini-3.5-flash-oh-pptx-reference-formatting | null | null | null | null | null | Help me detect all dangling paper titles in the slides `/root/Awesome-Agent-Papers.pptx`, and do the following
- change the font type to Arial, font size to 16, font color to #989596, and disable bold if any
- adjust the box width, so that each title in displayed in one line
- put the title at the bottom center of each... | pptx-reference-formatting | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/pptx-reference-formatting","passed":true,"reward":1.0,"task_image":"openhands-skillsbe... | Help me detect all dangling paper titles in the slides `/root/Awesome-Agent-Papers.pptx`, and do the following
- change the font type to Arial, font size to 16, font color to #989596, and disable bold if any
- adjust the box width, so that each title in displayed in one line
- put the title at the bottom center of each... | 38 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/pptx-reference-formatting.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 1,114,501 | 19,130 | 27,681 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-grid-dispatch-operator | null | null | null | null | null | You are a Grid Dispatch Operator. Given a power network snapshot, you want to decide the generators' dispatches. The dispatches should be economically efficient while satisfying the energy market requirement:
1. DC power balance at each bus
2. Generator and transmission limits
3. Spinning reserve requirements with stan... | grid-dispatch-operator | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/grid-dispatch-operator/critic_attempt_1/run_0a42d16c70... | You are a Grid Dispatch Operator. Given a power network snapshot, you want to decide the generators' dispatches. The dispatches should be economically efficient while satisfying the energy market requirement:
1. DC power balance at each bus
2. Generator and transmission limits
3. Spinning reserve requirements with stan... | 52 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/grid-dispatch-operator.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 1,637,707 | 11,643 | 31,215 |
skills-fireworks_ai-accounts-fireworks-models-minimax-m3-oh-exam-block-sequencing | null | null | null | null | null | You are helping a university registrar schedule final exams. The exams are already been grouped into blocks. Your job is to assign the blocks to available slots so that students have as few tightly packed exams as possible.
The task uses a prepared exam scheduling instance, which has 24 exam blocks and 24 time slots.... | exam-block-sequencing | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/exam-block-sequencing/critic_attempt_1/run_48a8e4ff337... | You are helping a university registrar schedule final exams. The exams are already been grouped into blocks. Your job is to assign the blocks to available slots so that students have as few tightly packed exams as possible.
The task uses a prepared exam scheduling instance, which has 24 exam blocks and 24 time slots.... | 58 | runs/skills_bench/vllm_local/minimax/minimax-M3/benchflow__skillsbench/fireworks_ai/accounts/fireworks/models/minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/exam-block-sequencing.tar.gz | minimax-m3_sdk_513653c_maxiter_100_N_sdk-logprobs-all | false | null | fireworks_ai/accounts/fireworks/models/minimax-m3 | null | null | 1,804,148 | 147,677 | 176,045 |
skills-gpt-5.5-2026-04-23-oh-tictoc-unnecessary-abort-detection | null | null | null | null | null | The famous TicToc concurrency-control mechanism can sometimes abort transactions that did not actually need to abort and could have been left in execution. Your task is to identify the policy of detecting every unnecessary abort.
You are given a trace from a TicToc workload to test if you can detect unnecessary aborts... | tictoc-unnecessary-abort-detection | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/tictoc-unnecessary-abort-detection","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/t... | The famous TicToc concurrency-control mechanism can sometimes abort transactions that did not actually need to abort and could have been left in execution. Your task is to identify the policy of detecting every unnecessary abort.
You are given a trace from a TicToc workload to test if you can detect unnecessary aborts... | 23 | runs/skills_bench/openai/gpt-5.5/benchflow__skillsbench/gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/tictoc-unnecessary-abort-detection.tar.gz | gpt-5.5-2026-04-23_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gpt-5.5-2026-04-23 | null | null | 510,159 | 9,298 | 15,641 |
skills-gemini-gemini-3.5-flash-oh-fix-erlang-ssh-cve | null | null | null | null | null | Erlang/OTP SSH is the built-in SSH server component of the Erlang/OTP platform.
A critical vulnerability was discovered in the Erlang/OTP SSH server, allowing attackers to execute arbitrary system commands remotely without authentication by crafting specific SSH protocol messages. In `/app/workspace/otp_src_27.3.2` we... | fix-erlang-ssh-cve | null | null | null | null | null | null | null | null | null | null | skillsbench | null | {"conversation_persistence_dir":"/zfs1/users/brendan/benchmarks/eval_outputs/runs/skillsbench_valid43/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/raw_conversations/fix-erlang-ssh-cve","passed":true,"reward":1.0,"task_image":"openhands-skillsbench/fix... | Erlang/OTP SSH is the built-in SSH server component of the Erlang/OTP platform.
A critical vulnerability was discovered in the Erlang/OTP SSH server, allowing attackers to execute arbitrary system commands remotely without authentication by crafting specific SSH protocol messages. In `/app/workspace/otp_src_27.3.2` we... | 57 | runs/skills_bench/google/gemini-3.5-flash/benchflow__skillsbench/gemini/gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all/conversations/fix-erlang-ssh-cve.tar.gz | gemini-3.5-flash_sdk_513653c_maxiter_100_N_sdk-logprobs-all | true | null | gemini/gemini-3.5-flash | null | null | 2,425,392 | 21,902 | 26,689 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.