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

#ModelScore
1Claude Fable 5 (Max)41.79
2Claude Opus 535.04
3Claude Opus 4.8 (High)33.84
4Claude Opus 4.8 (Max)32.9
5Kimi K331.96
6GLM-5.2 (Max)31.7
7Claude Opus 4.7 (xHigh)28.56
8GPT-5.5 (xHigh)27.23
9Grok 4.5 (High)23.45
10Gemini 3.1 Pro (Preview)21.99
11GPT-5.4 (High)19

Interactive version: theaggregate.ai/benchmark?slug=posttrainbench · How It Works · Data refreshed daily, snapshot 2026-09-05.