benchgap
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 modelReact Native EvalsSWE-bench VerifiedSource
Composer 2 Fast94.9%83.2%estimated ± 5.0 pp, low confidence
Gemma 4 31B75.2%76.1%estimated ± 5.0 pp, medium confidence

Fitted on

ModelReact Native EvalsSWE-bench Verified
Claude Opus 4.684.1%80.8%
Claude Opus 4.782.8%87.6%
Claude Sonnet 4.680.6%79.6%
GLM-574.8%77.8%
GPT-OSS 120B71.6%62.4%
GPT-OSS 20B71.0%60.7%
Kimi K2.577.2%76.8%