Calibration
Toolathlon → τ³-bench results
τ³-bench results is estimated from Toolathlon with a linear curve fitted on 6 models measured on both: y = 0.2161·x + 0.6032, R² = 0.43, cross-validated error 2.1 pp. It is used for 5 estimates.
| Estimated model | Toolathlon | τ³-bench results | Source |
|---|---|---|---|
| GPT-5.6 Luna | 53.4% | 71.9% | estimated ± 2.1 pp, low confidence |
| GPT-5.6 Sol | 58.0% | 72.9% | estimated ± 2.1 pp, low confidence |
| GPT-5.6 Terra | 53.1% | 71.8% | estimated ± 2.1 pp, low confidence |
| MiniMax M2.7 | 46.3% | 70.3% | estimated ± 2.1 pp, low confidence |
| Step 3.7 Flash | 49.5% | 71.0% | estimated ± 2.1 pp, low confidence |