τ-Bench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
τ-Bench
62.58Score
100
τ²-bench
72.4ASR
50
τ-Bench
85.5Average Pass@1
45
∞ Bench
90.39Accuracy
32
τ-bench-retail
70.2Success Rate
31
τ-BENCH (test)
0.791Average Task Reward
27
τ-Bench Retail (test)
94.5User Profile Success Rate
19
τ-bench Average
40Success Rate
18
τ²-Bench
75.5Airline Success Rate
18
τ-Bench and τ2-Bench (test)
82.6GED Score
18
τ²-Bench Telecom
100Accuracy
18
∞-Bench Longbook QA English (test)
11.2F1 Score
18
τ2-Bench
79.4Retail Performance
16
τ-bench retail (public leaderboard)
80.9Score
14
τ-Bench Retail
94.5User Profile Adherence
14
τ-Bench Airline
98.7User Profile Alignment (Prof.)
14
τ-Bench
63.9Avg Accuracy
14
τ-Bench (TauB) V2
91.6Accuracy
13
∞Bench
78.6MC Accuracy
13
∞Bench
78.46Accuracy
13
τ²-BENCH airline full 50-task pool
78Pass@4 Success Rate
12
τ-bench retail domain
28.7Action Reward
12
τ-bench airline
76Pass@1
12
∞Bench En.MC
90Accuracy
12
τ-Bench Telecom Verified (test)
94Pass@1
11