Calibration
OmniDocBench 1.5 → We-Math
We-Math is estimated from OmniDocBench 1.5 with a Hill curve fitted on 5 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.77803^6.00 + x^6.00), R² = 0.32, cross-validated error 6.9 pp. It is used for 0 estimates.
Fitted on
| Model | OmniDocBench 1.5 | We-Math |
|---|---|---|
| Claude Opus 4.5 | 87.7% | 70.0% |
| Gemini 3 Pro | 88.5% | 86.9% |
| GPT-5.2 | 85.7% | 79.0% |
| Kimi K2.5 | 88.8% | 84.7% |
| Qwen3.5 397B | 90.8% | 87.9% |