Calibration
SciCode → AA Coding Index
AA Coding Index is estimated from SciCode with a linear curve fitted on 15 models measured on both: y = 1.7307·x + -0.2911, R² = 0.76, cross-validated error 9.2 pp. It is used for 1 estimate.
| Estimated model | SciCode | AA Coding Index | Source |
|---|---|---|---|
| Ling 3.0 Flash FP8 | 40.4% | 40.8% | estimated ± 9.2 pp, medium confidence |
Fitted on
| Model | SciCode | AA Coding Index |
|---|---|---|
| Gemini 3.5 Flash | 53.1% | 70.1% |
| Granite 4.2 8B | 36.1% | 22.4% |
| Grok 4.3 | 47.3% | 42.3% |
| Hy3 Preview | 41.2% | 58.8% |
| Inkling-Small | 48.7% | 52.9% |
| Kimi K2.6 | 52.2% | 61.8% |
| Kimi K2.5 | 48.7% | 46.8% |
| Ling 2.6 Flash | 27.0% | 25.3% |
| Ling 3.0 Flash | 41.2% | 50.6% |
| Muse Glimmer 30B | 43.6% | 49.0% |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 31.4% | 26.8% |
| Nemotron 3 Nano Omni 30B A3B | 32.0% | 13.8% |
| Nemotron 3 Ultra | 44.6% | 49.3% |
| Qwen3.7 Max | 53.5% | 66.0% |
| Qwen3.7 Plus | 51.3% | 55.9% |