reasoning benchmark
Tau-bench
τ-bench: A benchmark for tool-agent-user interaction in real-world domains. Tests language agents' ability to interact with users and follow domain-specific rules through dynamic conversations using API tools and policy guidelines across retail and airline domains. Evaluates consistency and reliability of agent behavior over multiple trials.
Updated Aug 7, 2026
Published models6
Registry coverage6
MetricScore
EvidenceB
Tau-bench leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | ST | 88.2% | 100.0% | 6 | C | |
| 2 | ZA | 87.4% | 80.0% | 6 | C | |
| 3 | XI | 80.3% | 60.0% | 6 | C | |
| 4 | ZA | 79.5% | 40.0% | 6 | C | |
| 5 | MI | 77.2% | 20.0% | 6 | C | |
| 6 | OP | 63.0% | 0.0% | 6 | C |
Score distribution
Top published rows on the benchmark's original scale.
Tau-bench highlights
The top published results on this benchmark's own scale.
What is Tau-bench?
Definition and scoring fields from the benchmark registry.
τ-bench: A benchmark for tool-agent-user interaction in real-world domains. Tests language agents' ability to interact with users and follow domain-specific rules through dynamic conversations using API tools and policy guidelines across retail and airline domains. Evaluates consistency and reliability of agent behavior over multiple trials.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- Tau-bench
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- tau-bench|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about Tau-bench.
Which model scores highest on Tau-bench?
Step-3.5-Flash is currently ranked first with 88.2%.
What does Tau-bench measure?
τ-bench: A benchmark for tool-agent-user interaction in real-world domains. Tests language agents' ability to interact with users and follow domain-specific rules through dynamic conversations using API tools and policy guidelines across retail and airline domains. Evaluates consistency and reliability of agent behavior over multiple trials.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
6 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.