Skip to content

Benchmark

tau2-bench v0.1 retail pass^2

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models measured
6
Scores recorded
7
Sources
1
IQ weight
0.82

Score distribution

64%
Mean
66.8%
Median
7%
Std dev
58.7%
10th percentile
69.4%
90th percentile
49.8%
Min
49.8%71.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.82 for chat models, measured over 6 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT 5submission-gpt-5_sierra_2025-08-09, usersim-gpt-4.1-2025-04-14· Sierra tau-bench Leaderboard data docs71.8%
2Qwen3 Maxsubmission-qwen3-max_qwen_2026-01-23, usersim-gpt-4.1-2025-04-14· Sierra tau-bench Leaderboard data docs69.4%
3Qwen3 Maxsubmission-qwen3-max_qwen_2024_09_23· Sierra tau-bench Leaderboard data docs67.5%
4Claude Sonnet 3.7submission-claude-3-7-sonnet_anthropic_2024-06-20, usersim-gpt-4.1-2025-04-14· Sierra tau-bench Leaderboard data docs66.8%
5GPT 4.1submission-gpt-4-1_openai_2024-06-20, usersim-gpt-4.1-2025-04-14· Sierra tau-bench Leaderboard data docs64.2%
6o4 Minisubmission-o4-mini_openai_2024-06-20, usersim-gpt-4.1-2025-04-14· Sierra tau-bench Leaderboard data docs58.7%
7GPT-4.1 minisubmission-gpt-4-1-mini_openai_2024-06-20, usersim-gpt-4.1-2025-04-14· Sierra tau-bench Leaderboard data docs49.8%

Data credits: Sierra Research tau2-bench leaderboard submissions (github.com/sierra-research/tau2-bench), MIT (licence)