Agent benchmarks

τ³-Banking Knowledge agent benchmark

OpenGPT keeps 21 published rows attached to the exact τ³-Banking Knowledge benchmark version and environment shown below.

Tasks: 97
Published rows: 21
Source data date: Aug 4, 2026
Identity reviewed: Current

Evidence scope

v1.0.1 · AllTools · GPT-5.2 low user simulator · seed 300 · 4 trials. banking knowledge tasks. Standard scaffold with BM25, embedding retrieval, and sandboxed command-line search. 4 trials per task · 388 trajectories per configuration.

How OpenGPT handles this evidence

OpenGPT preserves the publisher's metric, task scope, environment, version, configuration fields, and source date. It does not create a cross-benchmark overall Agent score.

Comparison boundaries

A higher result applies only to the displayed benchmark version and conditions.
Agent results reflect the full system, not the base model alone.
Missing cost, reliability, safety, or latency data remains unreported rather than being inferred.
The included rows use the same benchmark version, 97-task Banking Knowledge split, AllTools configuration, user simulator, seed, and four-trial protocol. Missing costs remain unreported. This snapshot includes all 21 compatible publisher submissions available at the pinned revision.

v1.0.1 · AllTools · GPT-5.2 low user simulator · seed 300 · 4 trials

Showing 21 of 21 published rows. Every result remains attached to the source benchmark and configuration.