benchgap
Calibration

AA Coding Index → FrontierCode 1.1 Extended

FrontierCode 1.1 Extended is estimated from AA Coding Index with a Michaelis–Menten + offset curve fitted on 6 models measured on both: y = 0.0000 + 2.0000·x / (1.77063 + x), R² = 0.36, cross-validated error 3.4 pp. It is used for 3 estimates.

Estimated modelAA Coding IndexFrontierCode 1.1 ExtendedSource
Gemini 3.7 Flash76.1%60.1%estimated ± 3.4 pp, low confidence
Qwen3.8 Max Preview71.8%57.7%estimated ± 3.4 pp, low confidence
Gemini 3.1 Pro68.8%56.0%estimated ± 3.4 pp, low confidence

Fitted on

ModelAA Coding IndexFrontierCode 1.1 Extended
Claude Opus 578.0%63.6%
GPT-5.6 Luna71.5%55.1%
GPT-5.6 Sol77.4%60.6%
GPT-5.6 Terra76.7%55.8%
GPT-6 Astra76.9%64.5%
Grok 4.676.8%61.3%