Release planning · no model scores

tau2-bench benchmark evidence

How tau2-bench evidence maps to SpringPrompt V3 business-skill collections. This page publishes no model scores.

registered rights: approved source version: production-s3-banking-alltools-v1

Attribution and rights

tau2-bench

tau2-bench by Sierra Research and contributors (MIT License)

Publish only the attributed aggregate Pass^1 result and provenance. The large public trajectory conversations are used for verification but are not copied into SpringPrompt's public release.

Source homepage ↗Methodology ↗Attribution source ↗Terms ↗

Evidence hierarchy

1 mapped benchmark

tau2/tau3 support domains

1 V3 mapping · 1 selected metric field

planning only

Access: machine-ready
Registry: registered
Subject partitions: harnessed-model

Collections

Customer Support & Service Resolution

Capability atoms

knowledge_groundingsupport_resolution