Calibration
MCP-Tasks → MCP Atlas
MCP Atlas is estimated from MCP-Tasks with a inverse Michaelis–Menten curve fitted on 5 models measured on both: y = 0.35292·(x − 0.0000) / (0.0000 + 1.3016 − x), R² = 0.99, cross-validated error 1.5 pp. It is used for 0 estimates.
Fitted on
| Model | MCP-Tasks | MCP Atlas |
|---|---|---|
| Claude Opus 4.5 | 71.8% | 42.3% |
| GLM-5 | 60.8% | 31.1% |
| Kimi K2.5 | 59.1% | 29.5% |
| Qwen3.5 397B | 74.2% | 46.1% |
| Qwen3.6 Plus | 74.1% | 48.2% |