Agents · Benchmark profile
An end-to-end autonomous scientific research benchmark with 40 tasks across 10 scientific domains, where agents receive related literature and raw data, then attempt to rediscover the hidden target paper.
Data verified 27 Jul 2026 · Methodology 1.8.0
Benchmark score on ResearchClawBench
Claude Opus 4.8 leads at 21.1%, followed by Claude Opus 4.7 (20.7%) and GLM-5.2 (20.7%).
10 ranked models · higher is better · labels show rank and score
| Model | Rank | Provider | Score |
|---|---|---|---|
| Claude Opus 4.8 | #1 | Anthropic | 21.1% |
| Claude Opus 4.7 | #2 | Anthropic | 20.7% |
| GLM-5.2 | #3 | Z.AI | 20.7% |
| Claude Opus 4.6 | #4 | Anthropic | 19.9% |
| MiniMax M3 | #5 | MiniMax | 19.8% |
| Qwen3.7-Max | #6 | Alibaba Cloud | 18.7% |
| GLM-5.1 | #7 | Z.AI | 18.2% |
| Gemini 3.5 Flash | #8 | 18% | |
| Kimi K2.6 | #9 | Moonshot AI | 18% |
| Qwen3.6 Plus | #10 | Alibaba Cloud | 18% |
One best score per model · higher is better
| Rank | Model | Provider | License | Evidence use | Score |
|---|---|---|---|---|---|
| #1 | Claude Opus 4.8 claude-opus-4-8 | Anthropic | closed | Reference only | 21.1% |
| #2 | Claude Opus 4.7 claude-opus-4-7 | Anthropic | closed | Reference only | 20.7% |
| #3 | GLM-5.2 glm-5-2 | Z.AI | open | Estimated reference | 20.7% |
| #4 | Claude Opus 4.6 claude-opus-4-6 | Anthropic | closed | Reference only | 19.9% |
| #5 | MiniMax M3 MiniMax-M3 | MiniMax | open | Reference only | 19.8% |
| #6 | Qwen3.7-Max qwen3.7-max | Alibaba Cloud | closed | Reference only | 18.7% |
| #7 | GLM-5.1 glm-5-1 | Z.AI | open | Reference only | 18.2% |
| #8 | Gemini 3.5 Flash gemini-3.5-flash | closed | Estimated reference | 18% | |
| #9 | Kimi K2.6 kimi-k2.6 | Moonshot AI | open | Estimated reference | 18% |
| #10 | Qwen3.6 Plus qwen3-6-plus | Alibaba Cloud | closed | Reference only | 18% |
| #11 | DeepSeek V4 Pro deepseek-v4-pro | DeepSeek | open | Reference only | 17.1% |
| #12 | GPT-5.5 gpt-5.5 | OpenAI | closed | Estimated reference | 17% |
| #13 | MiMo-V2.5 mimo-v2.5 | Xiaomi | open | Estimated reference | 16.9% |
| #14 | GPT-5.4 gpt-5.4 | OpenAI | closed | Reference only | 15.3% |
| #15 | MiMo-V2-Pro mimo-v2-pro | Xiaomi | closed | Reference only | 15.3% |
| #16 | Qwen3.5 397B qwen3.5-397b-a17b | Alibaba Cloud | open | Estimated reference | 14.2% |
| #17 | Kimi K2.5 kimi-k2.5 | Moonshot AI | open | Reference only | 14% |
| #18 | Grok 4.1 grok-4-1 | xAI | closed | Estimated reference | 13.5% |
| #19 | Gemini 3.1 Pro gemini-3-1-pro | closed | Estimated reference | 13.3% | |
| #20 | Grok 4.3 grok-4.3 | xAI | closed | Reference only | 12.4% |
The top of this snapshot is led by Claude Opus 4.8 at 21.1%; third place is 0.40 points behind. The top-10 spread is 3.1 points.
About ResearchClawBench
An end-to-end autonomous scientific research benchmark with 40 tasks across 10 scientific domains, where agents receive related literature and raw data, then attempt to rediscover the hidden target paper. Results stay tied to the exact model variant and evaluation system. Multiple systems for the same model use the best published score on this page; overall Lumina scoring uses the median of ranking-eligible rows.
Open benchmark source ↗FAQ
An end-to-end autonomous scientific research benchmark with 40 tasks across 10 scientific domains, where agents receive related literature and raw data, then attempt to rediscover the hidden target paper.
Claude Opus 4.8 by Anthropic currently leads with 21.1%.
20 models in the LuminaBench cohort have a qualifying score on this benchmark.
No. This benchmark is display-only and does not enter the overall Lumina composite.
Related
