PostTrainBench
The agent post-trains a small model end to end on a GPU budget; scored on the resulting model.
Agent Index weight
4%
in the Long-horizon autonomy pillar
Models tested
21
from the benchmark’s leaderboard and reports
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 Supported Anthropic | 44.6% | Claude Code | — | — | |
| 2 | Claude Fable 5.2 Supported Anthropic | 44.6% | Claude Code | — | — | |
| 3 | Claude Opus 5.5 Supported Anthropic | 43.8% | Claude Code | — | — | |
| 4 | GPT-6 Astra Supported OpenAI | 41.9% | Codex CLI | — | — | |
| 5 | Gemini 4 Carbon Supported Google | 41.0% | OpenCode | — | — | |
| 6 | Gemini 4 Argon Supported Google | 40.4% | OpenCode | — | — | |
| 7 | GPT-6.1 Sol Estimated OpenAI | 38.8% | Codex CLI | — | — | |
| 8 | Claude Fable 5 Supported Anthropic | 37.3% | Claude Code | — | — | |
| 9 | GLM-5.3-Flash Supported Z.ai (Zhipu) | 36.8% | Claude Code | — | — | |
| 10 | Claude Opus 5 Supported Anthropic | 35.7% | Claude Code | — | — | |
| 11 | GLM-5.3 Supported Z.ai (Zhipu) | 34.5% | Claude Code | — | — | |
| 12 | Kimi K3 Supported Moonshot AI | 31.5% | Claude Code | — | — | |
| 13 | GPT-5.6 Sol Supported OpenAI | 31.0% | Codex CLI | — | — | |
| 14 | Claude Opus 4.8 Supported Anthropic | 31.0% | Claude Code | — | — | |
| 15 | GLM-5.2 Supported Z.ai (Zhipu) | 30.4% | Claude Code | — | — | |
| 16 | Claude Opus 4.7 Supported Anthropic | 24.3% | Claude Code | — | — | |
| 17 | Muse Wonder Estimated Meta | 24.2% | OpenCode | — | — | |
| 18 | GPT-5.5 Supported OpenAI | 23.9% | Codex CLI | — | — | |
| 19 | Grok 4.5 Estimated xAI | 22.6% | Cursor CLI | — | — | |
| 20 | Gemini 3.1 Pro Preview Supported Google | 18.1% | OpenCode | — | — | |
| 21 | GPT-5.4 Supported OpenAI | 17.8% | Codex CLI | — | — |
For models without a PostTrainBench result, the index estimates one from Epoch’s Capabilities Index using a logistic fit on 21 models (residual SD 0.3 on the logit scale). Estimates are labelled and widen the model’s range.