AutoRLBench 45 Issue / 19 Solution - Issue: leaderboard
Issue-diagnosis split (45 tasks) of Thinking-Space AutoRLBench 45-issue / 19-solution edition measuring whether agents isolate and verify root causes of RL training failures; pass rate (%).
Metric: Root cause found, 45 failed training runs (%). Source: thinking-space.github.io. Saturation forecast: Not forecast. 9 models tracked. Updated 21 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash | Max | 31.1 | #67 |
| 2 | GLM-5.3 | Max | 28.9 | #72 |
| 3 | GPT-6 Astra | Max | 26.7 | #10 |
| 4 | GPT-6.1 Sol | Max | 26.7 | #15 |
| 5 | Qwen 3.8 Max | Max | 20 | #63 |
| 6 | Grok 4.7 | Max | 20 | #51 |
| 7 | Claude Fable 5.1 | Max | 17.8 | #20 |
| 8 | Kimi K3.1 Pro | Max | 13.3 | #36 |
| 9 | Kimi K3 | Max | 4.4 | #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-45-issue-19-solution-issue · How It Works · Data refreshed daily, snapshot 2026-11-21.