Calibration
SWE-bench Pro → Vals LiveCodeBench
Vals LiveCodeBench is estimated from SWE-bench Pro with a Hill curve fitted on 32 models measured on both: y = 0.0000 + (0.8931 − 0.0000)·x^6.00 / (0.36733^6.00 + x^6.00), R² = 0.40, cross-validated error 4.7 pp. It is used for 19 estimates.
| Estimated model | SWE-bench Pro | Vals LiveCodeBench | Source |
|---|---|---|---|
| Claude Mythos 5 | 80.3% | 88.5% | estimated ± 4.7 pp, medium confidence |
| Ling 3.1 Flash | 68.7% | 87.3% | estimated ± 4.7 pp, medium confidence |
| Hy4 preview | 65.7% | 86.7% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.5-397B | 65.1% | 86.5% | estimated ± 4.7 pp, medium confidence |
| Claude Opus 4.7 (Adaptive) | 64.3% | 86.3% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.0-397B | 62.2% | 85.7% | estimated ± 4.7 pp, medium confidence |
| Atria Dawn Preview | 59.6% | 84.7% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.5-35B-A3B | 59.6% | 84.7% | estimated ± 4.7 pp, medium confidence |
| Laguna S 2.1 | 59.4% | 84.6% | estimated ± 4.7 pp, medium confidence |
| Step 3.7 Flash | 56.3% | 82.9% | estimated ± 4.7 pp, medium confidence |
| GPT-5.2 | 55.6% | 82.5% | estimated ± 4.7 pp, medium confidence |
| Muse Spark | 52.4% | 79.8% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.0-35B | 50.4% | 77.7% | estimated ± 4.7 pp, medium confidence |
| Laguna XS 2.1 | 47.6% | 73.7% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.5-9B | 47.5% | 73.6% | estimated ± 4.7 pp, medium confidence |
| Ornith-1.0-9B | 42.9% | 64.1% | estimated ± 4.7 pp, low confidence |
| LongCat-Flash-Lite-Sparse | 40.6% | 57.8% | estimated ± 4.7 pp, low confidence |
| LLaDA2.2-flash | 30.1% | 20.8% | estimated ± 4.7 pp, low confidence |
| MiniCPM5-2B | 14.4% | 0.3% | estimated ± 4.7 pp, low confidence |
Fitted on
| Model | SWE-bench Pro | Vals LiveCodeBench |
|---|---|---|
| Claude Fable 5 | 80.0% | 89.8% |
| Claude Fable 5.1 | 81.2% | 90.5% |
| Claude Opus 4.8 | 69.2% | 87.8% |
| Claude Opus 5 | 79.2% | 89.0% |
| Claude Sonnet 5 | 63.2% | 82.4% |
| DeepSeek V4 Flash 0731 | 52.6% | 87.3% |
| DeepSeek V4 Pro 0813 | 55.4% | 87.5% |
| Gemini 3.5 Flash | 55.1% | 87.6% |
| Gemini 3.5 Flash-Lite | 54.2% | 79.0% |
| GLM-5.1 | 58.4% | 81.4% |
| GLM-5.2 | 62.1% | 69.5% |
| GPT-5.3 Codex | 56.8% | 87.3% |
| GPT-5.5 | 58.6% | 85.3% |
| GPT-5.6 Sol | 64.6% | 82.6% |
| GPT-5.6 Terra | 63.4% | 85.9% |
| Grok 4.20 | 51.8% | 84.3% |
| Grok 4.5 | 64.7% | 87.4% |
| Inkling | 54.3% | 85.5% |
| Inkling-Small | 55.9% | 85.9% |
| Kimi K2.6 | 58.6% | 86.8% |
| Laguna M.1 | 49.2% | 68.1% |
| Laguna XS.2 | 46.3% | 67.8% |
| Ling 3.0 Flash | 56.6% | 84.0% |
| MiMo-V2.5 | 56.1% | 81.5% |
| MiMo-V2.5-Pro | 57.2% | 81.4% |
| MiniMax M2.7 | 56.2% | 79.9% |
| MiniMax M3 | 59.0% | 82.2% |
| Muse Spark 1.1 | 61.5% | 85.9% |
| Qwen3.6 Plus | 56.6% | 86.0% |
| Qwen3.7 Max | 60.6% | 87.1% |
| Qwen3.8-27B | 61.7% | 84.0% |
| Qwen3.8 Max | 67.7% | 87.9% |