Calibration
Toolathlon → AA ITBench
AA ITBench is estimated from Toolathlon with a inverse Michaelis–Menten curve fitted on 7 models measured on both: y = 1.24050·(x − 0.0000) / (0.0000 + 2.0000 − x), R² = 0.41, cross-validated error 6.1 pp. It is used for 0 estimates.
Fitted on
| Model | Toolathlon | AA ITBench |
|---|---|---|
| Gemini 3.5 Flash | 56.5% | 40.3% |
| GLM-5.2 | 48.2% | 42.7% |
| GPT-5.4 mini | 42.9% | 35.2% |
| GPT-5.5 | 55.6% | 45.8% |
| GPT-5.6 Luna | 53.4% | 40.3% |
| GPT-5.6 Sol | 58.0% | 56.2% |
| GPT-5.6 Terra | 53.1% | 51.0% |