τ-Bench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
τ-Bench Retail Verified (test)
92Pass Rate
11
τ-Bench Airline Verified (test)
82Pass@1
11
∞BENCH (test)
85.6Overall Accuracy
10
τ²-Bench Retail
90.4Accuracy
10
τ-Bench Average across Retail and Airline
18.1Success Rate (%)
8
τ-bench airline
19Success Rate
8
τ-Bench Retail, N=5
0.111Task Completion Rate
8
τ²-Bench (out-of-distribution)
54.9Retail Score
8
τ²-Bench telecom
98.7Score
7
∞Bench Zh.QA
49.1F1 Score
7
∞Bench En.QA
42.1F1 Score
7
τ-bench 157 scenarios
45.5Collaboration SR
7
τ-Bench (test)
60.4Retail Accuracy
7
τ²-bench telecom v1 (test)
85.9Pass@1
6
τ²-bench airline v1 (test)
53.5Pass@1
6
τ-bench
0.71Spearman Correlation (Rs)
6
τ-bench retail (test)
57.5Pass@4 Success Rate
6
τ-bench retail
44.173Pass@1 Success Rate
6
τ-bench
76.52Retail Success Rate
6
τ-bench airline multi-step
100Success Rate
5
τ-bench airline (80/20 stratified split)
89Top-1 Accuracy
5
τ²-Bench Airline
67.5Accuracy
5
τ²-Bench
76.6Accuracy
5
τ-bench
6.9Behavioral Uncertainty (BU)
5
τ-bench
38.3τ-bench Score
4