Calibration
AA EnterpriseOps-Gym → τ²-bench results
τ²-bench results is estimated from AA EnterpriseOps-Gym with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.2017·x / (0.12138 + x), R² = 0.94, cross-validated error 2.4 pp. It is used for 8 estimates.
| Estimated model | AA EnterpriseOps-Gym | τ²-bench results | Source |
|---|---|---|---|
| GLM-5.3 | 36.4% | 90.1% | estimated ± 2.4 pp, medium confidence |
| GLM-5.3-Flash | 33.2% | 88.0% | estimated ± 2.4 pp, medium confidence |
| Grok 4.6 | 48.3% | 96.0% | estimated ± 2.4 pp, medium confidence |
| Inkling | 38.0% | 91.1% | estimated ± 2.4 pp, medium confidence |
| Kimi K3 | 45.3% | 94.8% | estimated ± 2.4 pp, medium confidence |
| Muse Glimmer 30B | 34.7% | 89.0% | estimated ± 2.4 pp, medium confidence |
| Qwen3.8-27B | 44.2% | 94.3% | estimated ± 2.4 pp, medium confidence |
| Step 5 Preview | 47.2% | 95.6% | estimated ± 2.4 pp, medium confidence |