Datasets:

nebius-build-week
/

tryouts-eval

---
license: cc-by-4.0
language:
- en
- de
pretty_name: "Tryouts - which open model should my research agent use?"
task_categories:
- question-answering
tags:
- agents
- deep-research
- tool-use
- mcp
- evaluation
- nebius
- tavily
- nemotron
size_categories:
- n<1K
configs:
- config_name: hard_multihop
  data_files: data/hard_multihop.jsonl
- config_name: own_logs
  data_files: data/own_logs.jsonl
- config_name: tool_use
  data_files: data/tool_use.jsonl
- config_name: single_facts
  data_files: data/single_facts.jsonl
- config_name: results_hard_multihop
  data_files: results/hard_multihop.jsonl
- config_name: results_hard_multihop_run2
  data_files: results/hard_multihop_run2.jsonl
- config_name: results_own_logs
  data_files: results/own_logs.jsonl
- config_name: results_own_logs_deep
  data_files: results/own_logs_deep.jsonl
- config_name: results_tool_use
  data_files: results/tool_use.jsonl
- config_name: results_single_facts
  data_files: results/single_facts.jsonl
---
# Tryouts

Evaluation sets and full run logs for one question: **which open model should a web research agent use?**
The same agent (same prompts, same Tavily search) ran across the open models on
[Nebius Token Factory](https://tokenfactory.nebius.com) and Claude Opus 5 as the baseline.
Code to rerun everything on your own questions: https://github.com/Malek1414/tryouts.
Leaderboard: https://huggingface.co/spaces/nebius-build-week/tryouts.

Built at the Nebius Build Week hackathon with NVIDIA, Berlin, 18 October 2026, by team Starke Ameisen.

## Sets

| Config | Items | What it tests |
|---|---:|---|
| `single_facts` | 50 | Verified single facts in German and English. Every model answers from the same Tavily results. |
| `hard_multihop` | 40 | Questions from [FRAMES](https://huggingface.co/datasets/google/frames-benchmark) (numerical, temporal, multiple constraints) that chain 2 to 5 facts. Each model plans its own searches, up to 3 rounds. |
| `own_logs` | 16 | Written from research our own agents ran in the last month (Nightcrawler and Searchify). Each answer is backed by a verbatim quote checked against the page, and some questions rest on a false premise, so the right answer is "unknown". |
| `own_logs_deep` (results only) | 16 | The own-logs questions in deep mode: full pages, Searchify's reranker (`cross-encoder/mmarco-mMiniLMv2-L12-H384-v1`), one query per fact and a quote gate. |
| `tool_use` | 14 | The model drives real MCP servers through tool calls: Tavily search, extract, map and crawl; semble code search on GitHub; a headless Chrome for pages that only render with JavaScript. Gold answers were checked against the source on 18 October 2026. |

Fields: `question`, `answer` (gold), `aliases`, `answer_type` (`exact`, `number`, `unanswerable`), `source`, and for
`tool_use` the intended `tool`. The `results_*` configs hold every run: answer, verdict, search queries or tool calls,
sources, tokens, cost in USD and seconds.

The exact Tavily results every model read are in `search_cache/tavily_cache.jsonl`: drop it into the repo's
`results/` folder and a rerun replays the same searches without spending credits.

## Scoring

Verdicts follow SimpleQA: `correct`, `incorrect` (a wrong answer given with confidence) and `not_attempted`
(the model said it could not find it). On `unanswerable` items only "unknown" is correct. Numbers allow 1% tolerance.
Anything a rule cannot settle goes to a fixed judge model (Kimi K2.6) that is not a contestant.
Accuracy intervals are bootstrap 95% intervals (10,000 resamples). The pick is the cheapest model whose accuracy is
not credibly more than 10 points below the best one.

## Results (18 October 2026)

<!-- results:start -->
| Model | Single facts | Hard multi-hop | Our own logs | Tool use (MCP) | $ / question (hard) |
|---|---:|---:|---:|---:|---:|
| Claude Opus 5 | 100% (100%–100%), 0% wrong | **88%** (78%–98%), 12% wrong | 69% (44%–88%), 19% wrong | 100% (100%–100%), 0% wrong | $0.0619 |
| Reflection Beam | 100% (100%–100%), 0% wrong | 80% (68%–92%), 12% wrong | 69% (44%–88%), 19% wrong | 86% (64%–100%), 7% wrong | $0.0047 |
| Nemotron 3 Super | 100% (100%–100%), 0% wrong | 77% (64%–90%), 13% wrong | 56% (31%–81%), 31% wrong | 86% (64%–100%), 7% wrong | $0.0054 |
| Nemotron 3.5 Super VL | – | 77% (64%–90%), 8% wrong | 75% (50%–94%), 19% wrong | 100% (100%–100%), 0% wrong | $0.0076 |
| DeepSeek V4 Flash | 100% (100%–100%), 0% wrong | 70% (55%–82%), 8% wrong | **75%** (50%–94%), 19% wrong | **100%** (100%–100%), 0% wrong | $0.0021 |
| Nemotron 3.5 Lightning | 96% (90%–100%), 4% wrong | 65% (50%–80%), 20% wrong | 69% (44%–88%), 25% wrong | 64% (36%–86%), 36% wrong | $0.0028 |
| Nemotron 3 Ultra | 100% (100%–100%), 0% wrong | 57% (42%–72%), 15% wrong | 62% (38%–88%), 38% wrong | 0% (0%–0%), 100% wrong | $0.0244 |
| gpt-oss 120B | 96% (90%–100%), 4% wrong | 55% (40%–70%), 28% wrong | 56% (31%–81%), 44% wrong | 57% (29%–79%), 14% wrong | $0.0018 |
| Qwen3 30B A3B | **98%** (94%–100%), 2% wrong | 50% (35%–65%), 35% wrong | 62% (38%–88%), 31% wrong | 79% (57%–100%), 21% wrong | $0.0009 |
| Nemotron 3.5 Lightning (no thinking) | 94% (86%–100%), 6% wrong | 38% (22%–52%), 52% wrong | 56% (31%–81%), 44% wrong | – | $0.0005 |
| Nemotron 3 Nano | 96% (90%–100%), 4% wrong | – | – | 86% (64%–100%), 0% wrong | – |

Bold: the pick for that set (cheapest model not credibly more than 10 points below the best). Ranges: bootstrap 95% intervals. "Wrong" counts confident wrong answers, not "unknown".
<!-- results:end -->

Per-question replay: see the Space.

## Run-to-run variance (hard set, two passes)

<!-- variance:start -->
| Model | Run 1 | Run 2 | Questions that flipped |
|---|---:|---:|---:|
| Reflection Beam | 82% | 79% | 5 of 39 |
| Claude Opus 5 | 88% | 90% | 3 of 40 |
| Nemotron 3 Super | 77% | 62% | 8 of 39 |
| Qwen3 30B A3B | 49% | 41% | 9 of 39 |
<!-- variance:end -->

## Caveats

- Small sets: intervals are wide, read them before the point estimates.
- An early bug cached Tavily rate-limit replies as empty searches; affected runs were removed and rerun. See REPRODUCE.md in the repo.
- Nemotron 3 Ultra's tool calls came back from Token Factory as raw `<tool_call>` text or under invented tool names
  when 8 tools were offered (with one tool it called it correctly). This looks like a serving or chat-template issue,
  not the model's ability.
- Prices are Token Factory list prices on 18 October 2026.
- Facts with dates ("as of October 2026") can go stale.

## Licence

Our questions and run logs: CC BY 4.0. `hard_multihop` items come from FRAMES (Apache 2.0, Google).

	---
	license: cc-by-4.0
	language:
	- en
	- de
	pretty_name: "Tryouts - which open model should my research agent use?"
	task_categories:
	- question-answering
	tags:
	- agents
	- deep-research
	- tool-use
	- mcp
	- evaluation
	- nebius
	- tavily
	- nemotron
	size_categories:
	- n<1K
	configs:
	- config_name: hard_multihop
	data_files: data/hard_multihop.jsonl
	- config_name: own_logs
	data_files: data/own_logs.jsonl
	- config_name: tool_use
	data_files: data/tool_use.jsonl
	- config_name: single_facts
	data_files: data/single_facts.jsonl
	- config_name: results_hard_multihop
	data_files: results/hard_multihop.jsonl
	- config_name: results_hard_multihop_run2
	data_files: results/hard_multihop_run2.jsonl
	- config_name: results_own_logs
	data_files: results/own_logs.jsonl
	- config_name: results_own_logs_deep
	data_files: results/own_logs_deep.jsonl
	- config_name: results_tool_use
	data_files: results/tool_use.jsonl
	- config_name: results_single_facts
	data_files: results/single_facts.jsonl
	---

	# Tryouts

	Evaluation sets and full run logs for one question: which open model should a web research agent use?
	The same agent (same prompts, same Tavily search) ran across the open models on
	[Nebius Token Factory](https://tokenfactory.nebius.com) and Claude Opus 5 as the baseline.
	Code to rerun everything on your own questions: https://github.com/Malek1414/tryouts.
	Leaderboard: https://huggingface.co/spaces/nebius-build-week/tryouts.

	Built at the Nebius Build Week hackathon with NVIDIA, Berlin, 18 October 2026, by team Starke Ameisen.

	## Sets

	\| Config \| Items \| What it tests \|
	\|---\|---:\|---\|
	\| `single_facts` \| 50 \| Verified single facts in German and English. Every model answers from the same Tavily results. \|
	\| `hard_multihop` \| 40 \| Questions from [FRAMES](https://huggingface.co/datasets/google/frames-benchmark) (numerical, temporal, multiple constraints) that chain 2 to 5 facts. Each model plans its own searches, up to 3 rounds. \|
	\| `own_logs` \| 16 \| Written from research our own agents ran in the last month (Nightcrawler and Searchify). Each answer is backed by a verbatim quote checked against the page, and some questions rest on a false premise, so the right answer is "unknown". \|
	\| `own_logs_deep` (results only) \| 16 \| The own-logs questions in deep mode: full pages, Searchify's reranker (`cross-encoder/mmarco-mMiniLMv2-L12-H384-v1`), one query per fact and a quote gate. \|
	\| `tool_use` \| 14 \| The model drives real MCP servers through tool calls: Tavily search, extract, map and crawl; semble code search on GitHub; a headless Chrome for pages that only render with JavaScript. Gold answers were checked against the source on 18 October 2026. \|

	Fields: `question`, `answer` (gold), `aliases`, `answer_type` (`exact`, `number`, `unanswerable`), `source`, and for
	`tool_use` the intended `tool`. The `results_*` configs hold every run: answer, verdict, search queries or tool calls,
	sources, tokens, cost in USD and seconds.

	The exact Tavily results every model read are in `search_cache/tavily_cache.jsonl`: drop it into the repo's
	`results/` folder and a rerun replays the same searches without spending credits.

	## Scoring

	Verdicts follow SimpleQA: `correct`, `incorrect` (a wrong answer given with confidence) and `not_attempted`
	(the model said it could not find it). On `unanswerable` items only "unknown" is correct. Numbers allow 1% tolerance.
	Anything a rule cannot settle goes to a fixed judge model (Kimi K2.6) that is not a contestant.
	Accuracy intervals are bootstrap 95% intervals (10,000 resamples). The pick is the cheapest model whose accuracy is
	not credibly more than 10 points below the best one.

	## Results (18 October 2026)

	<!-- results:start -->
	\| Model \| Single facts \| Hard multi-hop \| Our own logs \| Tool use (MCP) \| $ / question (hard) \|
	\|---\|---:\|---:\|---:\|---:\|---:\|
	\| Claude Opus 5 \| 100% (100%–100%), 0% wrong \| 88% (78%–98%), 12% wrong \| 69% (44%–88%), 19% wrong \| 100% (100%–100%), 0% wrong \| $0.0619 \|
	\| Reflection Beam \| 100% (100%–100%), 0% wrong \| 80% (68%–92%), 12% wrong \| 69% (44%–88%), 19% wrong \| 86% (64%–100%), 7% wrong \| $0.0047 \|
	\| Nemotron 3 Super \| 100% (100%–100%), 0% wrong \| 77% (64%–90%), 13% wrong \| 56% (31%–81%), 31% wrong \| 86% (64%–100%), 7% wrong \| $0.0054 \|
	\| Nemotron 3.5 Super VL \| – \| 77% (64%–90%), 8% wrong \| 75% (50%–94%), 19% wrong \| 100% (100%–100%), 0% wrong \| $0.0076 \|
	\| DeepSeek V4 Flash \| 100% (100%–100%), 0% wrong \| 70% (55%–82%), 8% wrong \| 75% (50%–94%), 19% wrong \| 100% (100%–100%), 0% wrong \| $0.0021 \|
	\| Nemotron 3.5 Lightning \| 96% (90%–100%), 4% wrong \| 65% (50%–80%), 20% wrong \| 69% (44%–88%), 25% wrong \| 64% (36%–86%), 36% wrong \| $0.0028 \|
	\| Nemotron 3 Ultra \| 100% (100%–100%), 0% wrong \| 57% (42%–72%), 15% wrong \| 62% (38%–88%), 38% wrong \| 0% (0%–0%), 100% wrong \| $0.0244 \|
	\| gpt-oss 120B \| 96% (90%–100%), 4% wrong \| 55% (40%–70%), 28% wrong \| 56% (31%–81%), 44% wrong \| 57% (29%–79%), 14% wrong \| $0.0018 \|
	\| Qwen3 30B A3B \| 98% (94%–100%), 2% wrong \| 50% (35%–65%), 35% wrong \| 62% (38%–88%), 31% wrong \| 79% (57%–100%), 21% wrong \| $0.0009 \|
	\| Nemotron 3.5 Lightning (no thinking) \| 94% (86%–100%), 6% wrong \| 38% (22%–52%), 52% wrong \| 56% (31%–81%), 44% wrong \| – \| $0.0005 \|
	\| Nemotron 3 Nano \| 96% (90%–100%), 4% wrong \| – \| – \| 86% (64%–100%), 0% wrong \| – \|

	Bold: the pick for that set (cheapest model not credibly more than 10 points below the best). Ranges: bootstrap 95% intervals. "Wrong" counts confident wrong answers, not "unknown".
	<!-- results:end -->

	Per-question replay: see the Space.

	## Run-to-run variance (hard set, two passes)

	<!-- variance:start -->
	\| Model \| Run 1 \| Run 2 \| Questions that flipped \|
	\|---\|---:\|---:\|---:\|
	\| Reflection Beam \| 82% \| 79% \| 5 of 39 \|
	\| Claude Opus 5 \| 88% \| 90% \| 3 of 40 \|
	\| Nemotron 3 Super \| 77% \| 62% \| 8 of 39 \|
	\| Qwen3 30B A3B \| 49% \| 41% \| 9 of 39 \|
	<!-- variance:end -->

	## Caveats

	- Small sets: intervals are wide, read them before the point estimates.
	- An early bug cached Tavily rate-limit replies as empty searches; affected runs were removed and rerun. See REPRODUCE.md in the repo.
	- Nemotron 3 Ultra's tool calls came back from Token Factory as raw `<tool_call>` text or under invented tool names
	when 8 tools were offered (with one tool it called it correctly). This looks like a serving or chat-template issue,
	not the model's ability.
	- Prices are Token Factory list prices on 18 October 2026.
	- Facts with dates ("as of October 2026") can go stale.

	## Licence

	Our questions and run logs: CC BY 4.0. `hard_multihop` items come from FRAMES (Apache 2.0, Google).