Calibration
Toolathlon-Verified → CyberGym
CyberGym is estimated from Toolathlon-Verified with a Hill curve fitted on 7 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.63437^6.00 + x^6.00), R² = 0.40, cross-validated error 6.0 pp. It is used for 3 estimates.
| Estimated model | Toolathlon-Verified | CyberGym | Source |
|---|---|---|---|
| Claude Sonnet 5.5 | 77.8% | 92.7% | estimated ± 6.0 pp, low confidence |
| Muse Spark Open | 72.8% | 83.5% | estimated ± 6.0 pp, low confidence |
| Ornith-1.5-397B | 71.2% | 80.0% | estimated ± 6.0 pp, low confidence |
Fitted on
| Model | Toolathlon-Verified | CyberGym |
|---|---|---|
| DeepSeek V4 Flash 0731 | 70.3% | 76.7% |
| DeepSeek V4 Pro 0813 | 74.1% | 83.3% |
| GLM-5.3 | 73.0% | 84.5% |
| Hy4 preview | 74.1% | 78.4% |
| MiMo-V2.6-Flash | 73.6% | 95.1% |
| MiMo-V2.6-Pro | 76.9% | 94.0% |
| Step 5 Preview | 74.1% | 84.7% |