τ³-bench (tau3-bench)

Measures: An agent serves a simulated customer over several turns, using tools and following domain policy in the airline, retail, telecom and banking-knowledge domains (banking requires searching about 700 internal documents), plus a full-duplex voice track; scored as Pass^1, the share of tasks where the final database state and required actions match the goal.

Maintainer: Sierra (sierra-research)
Introduced: 2026
Status: Active
Saturation: Headroom
Typical provenance: Mixed
In scoring set: Yes

Paper Official leaderboard

Leaderboard

Show independent results only

Every result we hold, whoever ran it. Only independent and third-party rows count towards a score.

No results to show.