arXivVerified

BL-CLM-14.

τ-bench evaluates tool-using agents against goal database states and tests consistency across repeated trials under domain policies.

Claim visualization

Actual claim data

BL-CLM-14 / Relationship

Claim evidence map

arXiv
τ-bench evaluates tool-using agents against goal database states and tests consistency across repeated trials under domain policies.

The 2024 benchmark uses simulated users and domain environments. Its results are not Papyrus results or a measure of current models.

Source
arxiv.org/abs/2406.12045v1 ↗
Where in the source
v1 abstract; task evaluation and pass^k definition
Source organization
arXiv
Evidence type
Research benchmark
Source date
2024-06-17
Last verified
2026-10-09
Approved wording
τ-bench evaluates tool-using agents against goal database states and tests consistency across repeated trials under domain policies.
Scope and caveats
The 2024 benchmark uses simulated users and domain environments. Its results are not Papyrus results or a measure of current models.
Do not infer
Do not extrapolate benchmark scores into Papyrus performance or customer reliability.
Cited onAgentic Back Office ↗Modernization Factory ↗
← All claimsRecord BL-CLM-14