chat benchmark
TAU-bench Retail evaluates language agents' tool-agent-user interaction in retail environments through multi-turn conversations using domain-specific API tools and policy guidelines.
Updated Sep 5, 2026
Higher score ranks better on this benchmark.
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelAN | Score86.20% | Percentile100.00% | Participants25 | EvidenceC | Evaluated |
| Rank02 | ModelAN | Score82.40% | Percentile95.83% | Participants25 | EvidenceC | Evaluated |
| Rank03 | ModelAN | Score81.40% | Percentile91.67% | Participants25 | EvidenceC | Evaluated |
| Rank04 | ModelAN | Score81.20% | Percentile87.50% | Participants25 | EvidenceC | Evaluated |
| Rank05 | ModelAN | Score80.50% | Percentile83.33% | Participants25 | EvidenceC | Evaluated |
| Rank06 | ModelZA | Score79.70% | Percentile79.17% | Participants25 | EvidenceC | Evaluated |
| Rank07 | ModelZA | Score77.90% | Percentile75.00% | Participants25 | EvidenceC | Evaluated |
| Rank08 | ModelAC | Score77.50% | Percentile70.83% | Participants25 | EvidenceC | Evaluated |
| Rank09 | ModelOP | Score71.80% | Percentile66.67% | Participants25 | EvidenceC | Evaluated |
| Rank10 | ModelOP | Score70.80% | Percentile62.50% | Participants25 | EvidenceC | Evaluated |
| Rank11 | ModelAC | Score69.60% | Percentile58.33% | Participants25 | EvidenceC | Evaluated |
| Rank12 | ModelAN | Score69.20% | Percentile54.17% | Participants25 | EvidenceC | Evaluated |
| Rank13 | ModelOP | Score68.40% | Percentile50.00% | Participants25 | EvidenceC | Evaluated |
| Rank14 | ModelOP | Score68.00% | Percentile45.83% | Participants25 | EvidenceC | Evaluated |
| Rank15 | ModelOP | Score67.80% | Percentile41.67% | Participants25 | EvidenceC | Evaluated |
| Rank16 | ModelMI | Score67.80% | Percentile37.50% | Participants25 | EvidenceC | Evaluated |
| Rank17 | ModelAC | Score67.80% | Percentile33.33% | Participants25 | EvidenceC | Evaluated |
| Rank18 | ModelMI | Score63.50% | Percentile29.17% | Participants25 | EvidenceC | Evaluated |
| Rank19 | ModelAC | Score60.90% | Percentile25.00% | Participants25 | EvidenceC | Evaluated |
| Rank20 | ModelOP | Score60.30% | Percentile20.83% | Participants25 | EvidenceC | Evaluated |
| Rank21 | ModelOP | Score57.60% | Percentile16.67% | Participants25 | EvidenceC | Evaluated |
| Rank22 | ModelOP | Score55.80% | Percentile12.50% | Participants25 | EvidenceC | Evaluated |
| Rank23 | ModelOP | Score54.80% | Percentile8.33% | Participants25 | EvidenceC | Evaluated |
| Rank24 | ModelAN | Score51.00% | Percentile4.17% | Participants25 | EvidenceC | Evaluated |
| Rank25 | ModelOP | Score22.60% | Percentile0.00% | Participants25 | EvidenceC | Evaluated |
The leading models and scores on this benchmark.
A closer view of the leading scores on this benchmark.
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What TAU-bench Retail measures and how its scores work.
TAU-bench Retail is a benchmark for evaluating tool-agent-user interaction in retail environments.
It measures language agents' ability to handle dynamic multi-turn conversations with users, use domain-specific API tools, follow policy guidelines, and complete tasks such as order cancellations, address changes, and order status checks, reported as a Score ratio.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
Common questions about TAU-bench Retail.
Claude Sonnet 4.5 is currently ranked first with 86.20%.
The current leaders are Claude Sonnet 4.5 (86.20%), Claude Opus 4.1 (82.40%), and Claude Opus 4 (81.40%).
GPT-4.1 nano has the lowest matched official input price at $0.10 input / $0.40 output per 1M tokens.
The fastest matched records are GPT OSS 20B (1,000.00 tok/s via Groq), GPT OSS 120B (500.00 tok/s via Groq), and GPT-4.1 mini (467.39 tok/s via OpenAI).
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.
It measures language agents' ability to handle dynamic multi-turn conversations with users, use domain-specific API tools, follow policy guidelines, and complete tasks such as order cancellations, address changes, and order status checks, reported as a Score ratio.
Yes. Higher values rank better for this benchmark.
25 model results are currently shown.
No. This benchmark is shown for reference but does not contribute to the overall score.
Ranking basisThis tau-bench retail AI model leaderboard uses descending score in the benchmark original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
Selection summary
Claude Sonnet 4.5 currently leads TAU-bench Retail with 86.20%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price, and runtime.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.