Calibration
ProgramBench → CursorBench 3.2
CursorBench 3.2 is estimated from ProgramBench with a Michaelis–Menten curve fitted on 5 models measured on both: y = 2.0000·x / (1.66105 + x), R² = 0.92, cross-validated error 3.6 pp. It is used for 4 estimates.
| Estimated model | ProgramBench | CursorBench 3.2 | Source |
|---|---|---|---|
| DeepSeek V4.1 Flash | 20.3% | 21.8% | estimated ± 3.6 pp, low confidence |
| MiMo-V2.6-Flash | 26.0% | 27.1% | estimated ± 3.6 pp, low confidence |
| MiMo-V2.6-Pro | 26.5% | 27.5% | estimated ± 3.6 pp, low confidence |
| Step 5 Preview | 80.5% | 65.3% | estimated ± 3.6 pp, medium confidence |