Skip to content

PostTrainBench v1.1

Post-training four base language models across seven weighted benchmarks, with ten hours and one H100 per run.

15 models

Higher is better.

  1. 1Claude Opus 5.5Anthropic49.3
  2. 2Gemini 4 ArgonGoogle45.3
  3. 3GPT-6 AstraOpenAI44.3
  4. 4Claude Fable 5.1Anthropic40.2
  5. 5MiniMax M3.1 FlashMiniMax39.3
  6. 6GPT-5.6 SolOpenAI36.2
  7. 7Claude Opus 5Anthropic35.0
  8. 8Claude Opus 4.8Anthropic32.9
  9. 9Kimi K3Moonshot AI32.0
  10. 10GLM-5.2Z.AI31.7
  11. 11Claude Opus 4.7Anthropic28.6
  12. 12GPT-5.5OpenAI27.2
  13. 13Grok 4.5xAI23.4
  14. 14Gemini 3.1 ProGoogle22.0
  15. 15GPT-5.4OpenAI19.0
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Weighted aggregate across four base models and seven benchmarks
Tasks
Post-training Qwen3 1.7B and 4B, SmolLM3 3B, and Gemma3 4B
Authors
PostTrainBench team (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.