Leaderboard τ-bench (tau2-bench)
Customer-service agent simulations (airline, retail, telecom, banking, voice). Leaderboard data is in the tau2-bench repo.
- Operator
- Sierra
- Leaderboard
- taubench.com
- Category
- Agents / tool use
- How results are made
- Outside submissions
- Checked
- 2026-10-31
Raw data
| Format | Where |
|---|---|
| GitHub repo | github.com/sierra-research/tau2-bench Submissions arrive as pull requests; the leaderboard site lives in web/. “upload the trajectories to external storage, and open a pull request” (raw.githubusercontent.com) |
Licence / reuse: Reuse with attribution
MIT (repo) (licence of the repository that holds the result files). A licence or terms allow reuse if you credit the source (for example CC BY 4.0, MIT, Apache-2.0).
“MIT License”
Source: raw.githubusercontent.com/sierra-research/tau2-bench/main/LICENSE
Submissions: Open
Anyone can submit results, for example by pull request.
“1.x submissions are still accepted.”
Source: raw.githubusercontent.com/sierra-research/tau2-bench/main/docs/leaderboard-submission.md
Last updated: 2026-10-30 Updated in last 90 days
- 2026-10-30: Last commit touching web · github.com
This page lists where the data lives and what the licence says. It does not copy any scores; open the leaderboard for those. Not legal advice. Something wrong? Use the bug link in the footer. Back to all leaderboards.