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. Status: years away from saturation. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5 (Max) | 41.79 |
| 2 | Claude Opus 5 | 35.04 |
| 3 | Claude Opus 4.8 (High) | 33.84 |
| 4 | Claude Opus 4.8 (Max) | 32.9 |
| 5 | Kimi K3 | 31.96 |
| 6 | GLM-5.2 (Max) | 31.7 |
| 7 | Claude Opus 4.7 (xHigh) | 28.56 |
| 8 | GPT-5.5 (xHigh) | 27.23 |
| 9 | Grok 4.5 (High) | 23.45 |
| 10 | Gemini 3.1 Pro (Preview) | 21.99 |
| 11 | GPT-5.4 (High) | 19 |
Interactive version: theaggregate.ai/benchmark?slug=posttrainbench · How It Works · Data refreshed daily, snapshot 2026-09-05.