Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
schema_version: string
session_id: string
agent: struct<name: string, version: string, model_name: string, extra: struct<original_format: string, age (... 272 chars omitted)
child 0, name: string
child 1, version: string
child 2, model_name: string
child 3, extra: struct<original_format: string, agent_config: struct<system_template: string, instance_template: str (... 206 chars omitted)
child 0, original_format: string
child 1, agent_config: struct<system_template: string, instance_template: string, step_limit: int64, cost_limit: double, wa (... 159 chars omitted)
child 0, system_template: string
child 1, instance_template: string
child 2, step_limit: int64
child 3, cost_limit: double
child 4, wall_time_limit_seconds: int64
child 5, max_consecutive_format_errors: int64
child 6, output_path: string
child 7, mode: string
child 8, whitelist_actions: list<item: null>
child 0, item: null
child 9, confirm_exit: bool
steps: list<item: struct<step_id: int64, source: string, message: string>>
child 0, item: struct<step_id: int64, source: string, message: string>
child 0, step_id: int64
child 1, source: string
child 2, message: string
notes: string
final_metrics: struct<total_prompt_tokens: int64, total_completion_tokens: int64, total_steps: int64>
child 0, total_prompt_tokens: int64
child 1, total_completion_tokens: int64
child
...
tring
child 6, observation_template: string
child 7, multimodal_regex: string
child 8, action_regex: string
child 3, model_type: string
child 4, environment: struct<cwd: string, env: struct<PAGER: string, MANPAGER: string, LESS: string, PIP_PROGRESS_BAR: str (... 43 chars omitted)
child 0, cwd: string
child 1, env: struct<PAGER: string, MANPAGER: string, LESS: string, PIP_PROGRESS_BAR: string, TQDM_DISABLE: string (... 1 chars omitted)
child 0, PAGER: string
child 1, MANPAGER: string
child 2, LESS: string
child 3, PIP_PROGRESS_BAR: string
child 4, TQDM_DISABLE: string
child 2, timeout: int64
child 5, environment_type: string
child 2, mini_version: string
child 3, exit_status: string
child 4, submission: string
messages: list<item: struct<role: string, content: string, extra: struct<exit_status: string, submission: stri (... 47 chars omitted)
child 0, item: struct<role: string, content: string, extra: struct<exit_status: string, submission: string, excepti (... 35 chars omitted)
child 0, role: string
child 1, content: string
child 2, extra: struct<exit_status: string, submission: string, exception_str: string, traceback: string>
child 0, exit_status: string
child 1, submission: string
child 2, exception_str: string
child 3, traceback: string
trajectory_format: string
to
{'info': {'model_stats': {'instance_cost': Value('float64'), 'api_calls': Value('int64')}, 'config': {'agent': {'system_template': Value('string'), 'instance_template': Value('string'), 'step_limit': Value('int64'), 'cost_limit': Value('float64'), 'wall_time_limit_seconds': Value('int64'), 'max_consecutive_format_errors': Value('int64'), 'output_path': Value('string'), 'mode': Value('string'), 'whitelist_actions': List(Value('null')), 'confirm_exit': Value('bool')}, 'agent_type': Value('string'), 'model': {'model_name': Value('string'), 'model_kwargs': {'drop_params': Value('bool'), 'timeout': Value('int64'), 'temperature': Value('float64'), 'extra_body': {'reasoning_effort': Value('string')}, 'extra_headers': {'X-Session-ID': Value('string')}, 'max_tokens': Value('int64')}, 'litellm_model_registry': Value('null'), 'set_cache_control': Value('null'), 'cost_tracking': Value('string'), 'format_error_template': Value('string'), 'observation_template': Value('string'), 'multimodal_regex': Value('string'), 'action_regex': Value('string')}, 'model_type': Value('string'), 'environment': {'cwd': Value('string'), 'env': {'PAGER': Value('string'), 'MANPAGER': Value('string'), 'LESS': Value('string'), 'PIP_PROGRESS_BAR': Value('string'), 'TQDM_DISABLE': Value('string')}, 'timeout': Value('int64')}, 'environment_type': Value('string')}, 'mini_version': Value('string'), 'exit_status': Value('string'), 'submission': Value('string')}, 'messages': List({'role': Value('string'), 'content': Value('string'), 'extra': {'exit_status': Value('string'), 'submission': Value('string'), 'exception_str': Value('string'), 'traceback': Value('string')}}), 'trajectory_format': Value('string')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
schema_version: string
session_id: string
agent: struct<name: string, version: string, model_name: string, extra: struct<original_format: string, age (... 272 chars omitted)
child 0, name: string
child 1, version: string
child 2, model_name: string
child 3, extra: struct<original_format: string, agent_config: struct<system_template: string, instance_template: str (... 206 chars omitted)
child 0, original_format: string
child 1, agent_config: struct<system_template: string, instance_template: string, step_limit: int64, cost_limit: double, wa (... 159 chars omitted)
child 0, system_template: string
child 1, instance_template: string
child 2, step_limit: int64
child 3, cost_limit: double
child 4, wall_time_limit_seconds: int64
child 5, max_consecutive_format_errors: int64
child 6, output_path: string
child 7, mode: string
child 8, whitelist_actions: list<item: null>
child 0, item: null
child 9, confirm_exit: bool
steps: list<item: struct<step_id: int64, source: string, message: string>>
child 0, item: struct<step_id: int64, source: string, message: string>
child 0, step_id: int64
child 1, source: string
child 2, message: string
notes: string
final_metrics: struct<total_prompt_tokens: int64, total_completion_tokens: int64, total_steps: int64>
child 0, total_prompt_tokens: int64
child 1, total_completion_tokens: int64
child
...
tring
child 6, observation_template: string
child 7, multimodal_regex: string
child 8, action_regex: string
child 3, model_type: string
child 4, environment: struct<cwd: string, env: struct<PAGER: string, MANPAGER: string, LESS: string, PIP_PROGRESS_BAR: str (... 43 chars omitted)
child 0, cwd: string
child 1, env: struct<PAGER: string, MANPAGER: string, LESS: string, PIP_PROGRESS_BAR: string, TQDM_DISABLE: string (... 1 chars omitted)
child 0, PAGER: string
child 1, MANPAGER: string
child 2, LESS: string
child 3, PIP_PROGRESS_BAR: string
child 4, TQDM_DISABLE: string
child 2, timeout: int64
child 5, environment_type: string
child 2, mini_version: string
child 3, exit_status: string
child 4, submission: string
messages: list<item: struct<role: string, content: string, extra: struct<exit_status: string, submission: stri (... 47 chars omitted)
child 0, item: struct<role: string, content: string, extra: struct<exit_status: string, submission: string, excepti (... 35 chars omitted)
child 0, role: string
child 1, content: string
child 2, extra: struct<exit_status: string, submission: string, exception_str: string, traceback: string>
child 0, exit_status: string
child 1, submission: string
child 2, exception_str: string
child 3, traceback: string
trajectory_format: string
to
{'info': {'model_stats': {'instance_cost': Value('float64'), 'api_calls': Value('int64')}, 'config': {'agent': {'system_template': Value('string'), 'instance_template': Value('string'), 'step_limit': Value('int64'), 'cost_limit': Value('float64'), 'wall_time_limit_seconds': Value('int64'), 'max_consecutive_format_errors': Value('int64'), 'output_path': Value('string'), 'mode': Value('string'), 'whitelist_actions': List(Value('null')), 'confirm_exit': Value('bool')}, 'agent_type': Value('string'), 'model': {'model_name': Value('string'), 'model_kwargs': {'drop_params': Value('bool'), 'timeout': Value('int64'), 'temperature': Value('float64'), 'extra_body': {'reasoning_effort': Value('string')}, 'extra_headers': {'X-Session-ID': Value('string')}, 'max_tokens': Value('int64')}, 'litellm_model_registry': Value('null'), 'set_cache_control': Value('null'), 'cost_tracking': Value('string'), 'format_error_template': Value('string'), 'observation_template': Value('string'), 'multimodal_regex': Value('string'), 'action_regex': Value('string')}, 'model_type': Value('string'), 'environment': {'cwd': Value('string'), 'env': {'PAGER': Value('string'), 'MANPAGER': Value('string'), 'LESS': Value('string'), 'PIP_PROGRESS_BAR': Value('string'), 'TQDM_DISABLE': Value('string')}, 'timeout': Value('int64')}, 'environment_type': Value('string')}, 'mini_version': Value('string'), 'exit_status': Value('string'), 'submission': Value('string')}, 'messages': List({'role': Value('string'), 'content': Value('string'), 'extra': {'exit_status': Value('string'), 'submission': Value('string'), 'exception_str': Value('string'), 'traceback': Value('string')}}), 'trajectory_format': Value('string')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
routerbench trajectories
Full agent trajectories and graded results for routerbench, a study of task-level model routing for agentic coding.
Code, explanations, and current result tables: https://github.com/JerryYan123/routerbench
(start with docs/RESULTS.md). Prices behind every cost: docs/PRICES.md in the same repository.
Layout
| Path | Contents |
|---|---|
records/ |
One JSON line per (benchmark, task, model): resolved, tokens, cost_usd, wall time, steps, outcome, error, and traj_path (the trajectory below). Start here. |
swebench/ |
SWE-bench Verified runs with OpenHands (SDK e212d45). One directory per (model, subset). output.jsonl holds the full history of each task; conversations/ holds compressed event logs; logs/run_evaluation/ holds the official grader logs; output.report.json lists the solved tasks. |
claude_code/ |
SWE-bench Verified runs of Claude Opus 5.5 (effort xhigh) and Claude Fable 5.1 (effort high) in Claude Code, one directory per task: transcript, patch, grade log, result.json with the token split. |
swebench_pro/ |
SWE-bench Pro runs: open_models/ (OpenHands) and claude_and_gold/ (Claude Code runs and gold-patch checks). |
terminalbench/ |
Terminal-Bench 2.0 runs (Harbor + mini-swe-agent), one directory per trial with result.json and agent/trajectory.json. |
traj_path values in records/ are absolute paths on our cluster. Remove the prefix up to runs/ and map
runs/swebench → swebench/, runs/claude_code → claude_code/, runs/terminalbench → terminalbench/.
A path output.jsonl#<task id> means: the line of that file whose instance_id is <task id>.
Notes
- Records with a non-empty
errorare infrastructure failures (not model failures). They are excluded from results. - Task content (issues, repositories, tests) belongs to the original benchmarks and their licenses: SWE-bench Verified, SWE-bench Pro, Terminal-Bench 2.0. Terminal-Bench tasks carry a canary string; do not train on this data.
- Model outputs are research artifacts. API keys were removed before upload.
- Downloads last month
- 25