PostTrainBench Average (v1.1)
Indexes · 2026-03-09
Historical PostTrainBench v1.1 weighted average across seven downstream benchmarks and four base models. Superseded by v1.2, which removes BFCL, fixes HumanEval scoring, and averages final evaluations across seeds; retained for historical comparison only.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 49.3 |
| Claude Opus 5 | 45.6 |
| Gemini 4 Argon | 45.3 |
| GPT-6 Astra | 44.3 |
| Claude Fable 5.1 | 40.2 |
| MiniMax M3.1 Flash | 39.3 |
| GPT-5.6 Sol | 36.2 |
| Opus 4.8 | 33.8 |
| GLM-5.2 | 31.7 |
| Opus 4.7 | 28.6 |
| GPT-5.5 | 27.2 |
| Grok 4.5 | 23.4 |
| Gemini 3.1 Pro Preview | 22.0 |
| GPT-5.4 | 19.0 |