benchgap
Calibration

SWE-Rebench → SWE-bench Verified

SWE-bench Verified is estimated from SWE-Rebench with a Michaelis–Menten curve fitted on 8 models measured on both: y = 2.0000·x / (0.95233 + x), R² = 0.32, cross-validated error 4.8 pp. It is used for 3 estimates.

Estimated modelSWE-RebenchSWE-bench VerifiedSource
Composer 258.0%75.7%estimated ± 4.8 pp, medium confidence
DeepSeek V3.260.9%78.0%estimated ± 4.8 pp, medium confidence
Gemma 4 31B41.6%60.8%estimated ± 4.8 pp, low confidence