Agent: GPT-5.2 (xhigh)

Reliability Dimensions by Benchmark
τ-bench (airline, clean) Detail →
Accuracy 67.7%
Consistency 0.70
Predictability 0.78
Robustness 0.96
Safety 0.95
Reliability 0.81
τ-bench (airline, original) Detail →
Accuracy 51.6%
Consistency 0.67
Predictability 0.65
Robustness 0.95
Safety 0.94
Reliability 0.76