Benchmark

τ-Voice (tau-Voice)

Measures: A full-duplex voice agent handles simulated customer-service calls (airline, retail, telecom; banking on taubench.com) using tools and policy documents, while coping with interruptions, accents and background noise; a task passes if the final database state matches the ground truth (pass^1).

Maintainer
Sierra (tau-bench team)
Introduced
2026
Status
Active
Saturation
Headroom
Typical provenance
Mixed
In scoring set
Yes

PaperOfficial leaderboard

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

No results to show.

τ-Voice (tau-Voice): leaderboard and what it measures · supermarket.dev