ResearchClawBench: leaderboard
40 autonomous-research tasks in 10 scientific domains, each built from a published paper with its literature and data, scored on expert rubrics for re-discovering its findings; InternScience, 2026.
Metric: Overall (self-reported). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Around September 2027. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 20.7 | #49 |
| 2 | Claude Opus 4.6 | 19.9 | #63 |
| 3 | Qwen 3.7 Max | 18.7 | #75 |
| 4 | GLM-5.1 | 18.2 | #97 |
| 5 | Kimi K2.6 | 18 | #104 |
| 6 | Qwen 3.6 Plus | 18 | #120 |
| 7 | Gemini 3.5 Flash | 17.9 | #60 |
| 8 | DeepSeek V4 Pro | 17.1 | #102 |
| 9 | GPT-5.5 | 17 | #31 |
| 10 | MiMo-V2.5 | 16.9 | #196 |
| 11 | GPT-5.4 | 15.3 | #79 |
| 12 | MiMo-V2-Pro | 15.3 | #168 |
| 13 | Qwen 3.5 397B A17B | 14.2 | #146 |
| 14 | Kimi K2.5 | 14 | #144 |
| 15 | Grok 4.1 | 13.5 | #227 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=researchclawbench · How It Works · Data refreshed daily, snapshot 2026-10-20.