Calibration
Vals SWE-bench → FrontierCode 1.1 Main
FrontierCode 1.1 Main is estimated from Vals SWE-bench with a Hill curve fitted on 9 models measured on both: y = 0.0000 + (0.6981 − 0.0000)·x^6.00 / (0.78165^6.00 + x^6.00), R² = 0.80, cross-validated error 5.2 pp. It is used for 5 estimates.
| Estimated model | Vals SWE-bench | FrontierCode 1.1 Main | Source |
|---|---|---|---|
| Gemini 3 Flash | 75.0% | 30.6% | estimated ± 5.2 pp, medium confidence |
| Qwen 3.6 Max (preview) | 72.8% | 27.6% | estimated ± 5.2 pp, low confidence |
| MiMo-V2.5 | 71.0% | 25.1% | estimated ± 5.2 pp, low confidence |
| Qwen3.5 Flash | 64.4% | 16.6% | estimated ± 5.2 pp, low confidence |
| Gemini 3.1 Flash-Lite | 62.8% | 14.8% | estimated ± 5.2 pp, low confidence |
Fitted on
| Model | Vals SWE-bench | FrontierCode 1.1 Main |
|---|---|---|
| Claude Fable 5 | 95.0% | 53.5% |
| Claude Opus 4.7 | 82.0% | 38.5% |
| Claude Opus 4.8 | 88.6% | 46.5% |
| Claude Opus 5 | 97.0% | 53.4% |
| Claude Sonnet 4.6 | 77.4% | 24.3% |
| Claude Sonnet 5 | 79.6% | 42.7% |
| Gemini 3.7 Flash | 80.8% | 43.6% |
| GPT-5.4 mini | 73.0% | 27.0% |
| GPT-5.5 | 82.6% | 43.0% |