benchgap
Calibration

τ³-Bench Banking → EnterpriseOps-Gym

EnterpriseOps-Gym is estimated from τ³-Bench Banking with a Michaelis–Menten curve fitted on 10 models measured on both: y = 0.5171·x / (0.10531 + x), R² = 0.55, cross-validated error 5.3 pp. It is used for 6 estimates.

Estimated modelτ³-Bench BankingEnterpriseOps-GymSource
Gemini 3.8 Flash (high)44.9%41.9%estimated ± 5.3 pp, medium confidence
Muse Spark 1.3 (max)50.5%42.8%estimated ± 5.3 pp, medium confidence
K2 Horizon 375B A23B34.2%39.5%estimated ± 5.3 pp, medium confidence
Qwen3.8 Max51.3%42.9%estimated ± 5.3 pp, low confidence
Qwen3.8 2.4T A95B49.1%42.6%estimated ± 5.3 pp, medium confidence
Qwen3.8 27B (medium)47.4%42.3%estimated ± 5.3 pp, medium confidence