PostTrainBench: leaderboard
AI R&D automation benchmark: agents get 4 small LLMs, one H100, and 10 hours to improve model performance through post-training. Measures dataset curation, training strategy, and constraint awareness.
Metric: Weighted Avg Score. Source: posttrainbench.com. Saturation forecast: Around 2029. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 36.23 | #19 (GPT-5.6 Sol) |
| 2 | Claude Opus 5 | 35.04 | #12 |
| 3 | Claude Opus 4.8 (High) | 33.84 | #37 (Claude Opus 4.8) |
| 4 | Claude Opus 4.8 (Max) | 32.9 | #37 (Claude Opus 4.8) |
| 5 | Kimi K3 | 31.96 | #30 |
| 6 | GLM-5.2 (Max) | 31.7 | #73 (GLM-5.2) |
| 7 | Claude Opus 4.7 (xHigh) | 28.56 | #49 (Claude Opus 4.7) |
| 8 | GPT-5.5 (xHigh) | 27.23 | #31 (GPT-5.5) |
| 9 | Grok 4.5 (High) | 23.45 | #48 (Grok 4.5) |
| 10 | Gemini 3.1 Pro (Preview) | 21.99 | #58 |
| 11 | GPT-5.4 (High) | 19 | #79 (GPT-5.4) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=posttrainbench · How It Works · Data refreshed daily, snapshot 2026-10-20.