benchgap
Calibration

FrontierCode 1.1 Main → AA Coding Index

AA Coding Index is estimated from FrontierCode 1.1 Main with a Michaelis–Menten curve fitted on 9 models measured on both: y = 1.1662·x / (0.25924 + x), R² = 0.86, cross-validated error 4.4 pp. It is used for 1 estimate.

Estimated modelFrontierCode 1.1 MainAA Coding IndexSource
Claude Opus 4.626.9%59.4%estimated ± 4.4 pp, medium confidence

Fitted on

ModelFrontierCode 1.1 MainAA Coding Index
Claude Fable 553.5%76.5%
Claude Opus 4.738.5%73.6%
Claude Opus 4.846.5%74.3%
Claude Opus 553.4%78.0%
Claude Sonnet 542.7%71.6%
Gemini 3.7 Flash43.6%76.1%
GPT-5.4 mini27.0%56.1%
GPT-5.543.0%74.9%
GPT-6 Astra53.3%76.9%