Calibration
SWE-bench Pro → PostTrainBench v1.1
PostTrainBench v1.1 is estimated from SWE-bench Pro with a inverse Michaelis–Menten curve fitted on 9 models measured on both: y = 0.59838·(x − 0.0000) / (0.0000 + 2.0000 − x), R² = 0.75, cross-validated error 4.7 pp. It is used for 33 estimates.
| Estimated model | SWE-bench Pro | PostTrainBench v1.1 | Source |
|---|---|---|---|
| Claude Mythos 5 | 80.3% | 40.1% | estimated ± 4.7 pp, high confidence |
| Sakana Fugu-Ultra | 73.7% | 34.9% | estimated ± 4.7 pp, high confidence |
| Claude Opus 4.7 (Adaptive) | 64.3% | 28.4% | estimated ± 4.7 pp, high confidence |
| Ornith-1.0-397B | 62.2% | 27.0% | estimated ± 4.7 pp, high confidence |
| Sage-A1.5 | 61.2% | 26.4% | estimated ± 4.7 pp, high confidence |
| dots3-note Preview | 61.0% | 26.3% | estimated ± 4.7 pp, high confidence |
| Atria Dawn Preview | 59.6% | 25.4% | estimated ± 4.7 pp, high confidence |
| Ornith-1.5-35B-A3B | 59.6% | 25.4% | estimated ± 4.7 pp, high confidence |
| Sakana Fugu | 59.0% | 25.0% | estimated ± 4.7 pp, high confidence |
| Qwen3.7 Plus | 57.6% | 24.2% | estimated ± 4.7 pp, medium confidence |
| Claude Opus 4.5 | 57.1% | 23.9% | estimated ± 4.7 pp, medium confidence |
| Step 3.7 Flash | 56.3% | 23.4% | estimated ± 4.7 pp, medium confidence |
| GPT-5.2 | 55.6% | 23.0% | estimated ± 4.7 pp, medium confidence |
| GLM-5 | 55.1% | 22.8% | estimated ± 4.7 pp, medium confidence |
| Spark-X2.6-4B | 54.3% | 22.3% | estimated ± 4.7 pp, medium confidence |
| Claude Opus 4.6 | 53.4% | 21.8% | estimated ± 4.7 pp, medium confidence |
| MAI-Thinking-1 | 52.8% | 21.5% | estimated ± 4.7 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B | 51.8% | 20.9% | estimated ± 4.7 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B NVFP4 | 51.3% | 20.6% | estimated ± 4.7 pp, medium confidence |
| Muse Glimmer 30B | 51.2% | 20.6% | estimated ± 4.7 pp, medium confidence |
| Qwen3.5 397B | 50.9% | 20.4% | estimated ± 4.7 pp, medium confidence |
| Kimi K2.5 | 50.7% | 20.3% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.0-35B | 50.4% | 20.2% | estimated ± 4.7 pp, medium confidence |
| Qwen3.6-35B-A3B | 49.5% | 19.7% | estimated ± 4.7 pp, medium confidence |
| Laguna M.1 | 49.2% | 19.5% | estimated ± 4.7 pp, medium confidence |
| Laguna XS 2.1 | 47.6% | 18.7% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.5-9B | 47.5% | 18.6% | estimated ± 4.7 pp, medium confidence |
| Laguna XS.2 | 46.3% | 18.0% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.0-9B | 42.9% | 16.3% | estimated ± 4.7 pp, medium confidence |
| LongCat-Flash-Lite-Sparse | 40.6% | 15.3% | estimated ± 4.7 pp, medium confidence |
| Granite 4.2 30B | 33.3% | 11.9% | estimated ± 4.7 pp, medium confidence |
| LLaDA2.2-flash | 30.1% | 10.6% | estimated ± 4.7 pp, medium confidence |
| Mellum2.1-12B-A2.5B-Thinking | 28.0% | 9.7% | estimated ± 4.7 pp, medium confidence |
Fitted on
| Model | SWE-bench Pro | PostTrainBench v1.1 |
|---|---|---|
| Claude Fable 5.1 | 81.2% | 40.2% |
| Claude Opus 4.8 | 69.2% | 32.9% |
| Claude Opus 5 | 79.2% | 35.0% |
| Claude Opus 5.5 | 89.9% | 49.3% |
| GLM-5.2 | 62.1% | 31.7% |
| GPT-5.4 | 57.7% | 19.0% |
| GPT-5.5 | 58.6% | 27.2% |
| GPT-5.6 Sol | 64.6% | 36.2% |
| Grok 4.5 | 64.7% | 23.5% |