AutoRLBench 40 Issue / 20 Solution: leaderboard
Thinking-Space AutoRLBench 40-issue / 20-solution edition evaluating autonomous RL research agents end-to-end across 60 tasks (40 issue diagnosis and 20 solution implementation tasks) in sandboxed repositories; overall pass rate (%).
Metric: Pass rate over 40 issue judgments and 20 solution runs (%). Source: thinking-space.github.io. Saturation forecast: Around 2029. 6 models tracked. Updated 19 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | GPT-6.1 Sol | Max | 31.7 | #15 |
| 2 | GLM-5.3 | Max | 25 | #73 |
| 3 | Claude Fable 5.1 | Max | 18.3 | #20 |
| 4 | Qwen 3.8 Max | Max | 18.3 | #66 |
| 5 | Grok 4.7 | Max | 16.7 | #51 |
| 6 | Kimi K3 | Max | 6.7 | #61 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=autorlbench-40-issue-20-solution · How It Works · Data refreshed daily, snapshot 2026-11-20.