Calibration
React Native Evals → AA Coding Index
AA Coding Index is estimated from React Native Evals with a Hill curve fitted on 9 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.80210^6.00 + x^6.00), R² = 0.72, cross-validated error 11.1 pp. It is used for 3 estimates.
| Estimated model | React Native Evals | AA Coding Index | Source |
|---|---|---|---|
| Composer 2 | 96.1% | 89.7% | estimated ± 11.1 pp, low confidence |
| Composer 2 Fast | 94.9% | 87.9% | estimated ± 11.1 pp, low confidence |
| DeepSeek V3.2 | 71.5% | 40.1% | estimated ± 11.1 pp, low confidence |
Fitted on
| Model | React Native Evals | AA Coding Index |
|---|---|---|
| Claude Opus 4.7 | 82.8% | 73.6% |
| Gemini 3.1 Pro | 78.9% | 68.8% |
| Gemma 4 31B | 75.2% | 43.4% |
| GPT-5.4 | 85.3% | 71.1% |
| GPT-5.5 | 84.7% | 74.9% |
| GPT-OSS 120B | 71.6% | 30.4% |
| GPT-OSS 20B | 71.0% | 20.7% |
| Kimi K2.5 | 77.2% | 46.8% |
| MiniMax M2.7 | 71.4% | 52.6% |