Calibration
React Native Evals → SWE-bench Verified
SWE-bench Verified is estimated from React Native Evals with a offset logistic curve fitted on 7 models measured on both: y = 0.0000 + (0.8321 − 0.0000) / (1 + exp(−33.57·(x − 0.6812))), R² = 0.92, cross-validated error 5.0 pp. It is used for 2 estimates.
| Estimated model | React Native Evals | SWE-bench Verified | Source |
|---|---|---|---|
| Composer 2 Fast | 94.9% | 83.2% | estimated ± 5.0 pp, low confidence |
| Gemma 4 31B | 75.2% | 76.1% | estimated ± 5.0 pp, medium confidence |
Fitted on
| Model | React Native Evals | SWE-bench Verified |
|---|---|---|
| Claude Opus 4.6 | 84.1% | 80.8% |
| Claude Opus 4.7 | 82.8% | 87.6% |
| Claude Sonnet 4.6 | 80.6% | 79.6% |
| GLM-5 | 74.8% | 77.8% |
| GPT-OSS 120B | 71.6% | 62.4% |
| GPT-OSS 20B | 71.0% | 60.7% |
| Kimi K2.5 | 77.2% | 76.8% |