Project AURIONAI Use-Cases Research & Innovation

tau-bench

Tool-using agent talks to a simulated user in airline, retail and telecom domains

agents and tool usecrowded
2scores
1models
0independent
0maintainer-run
2vendor-reported

Definition

Version
not in the notes
Category
agents and tool use
Maintainer
Sierra
Status
crowded
Home page
https://arxiv.org/abs/2406.12045

Sources

Scores

2 of 2 scores shown tau-bench-scores.csv
ModelScoreWho ran itSettingView or populationDateSource
AnthropicClaude 3.5 Sonnet (Oct '24)69.2 %vendor-reportedretail domain–22 Oct 2024anthropic.com
AnthropicClaude 3.5 Sonnet (Oct '24)46 %vendor-reportedairline domain–22 Oct 2024anthropic.com
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Tau-Bench", https://aurion.semai.id/benchmarks/tau-bench/, records of October 2026.