Atlas

Benchmarks

← All benchmarks

BankerToolBench Pass Rate

Professional Work · 2026-04-13

Share of BankerToolBench v1.0's 100 tasks whose deliverable scores above the 0.80 rubric-score threshold that the authors calibrated against bankers' ratings of 'needs light edits' or 'sendable as-is', averaged over three runs (Pass@1). It is a component of the v1.0 factor group, derived from the same graded runs as the rubric Score.

Top models (higher is better)

ModelScore
GPT-5.416.0
Gemini 3.1 Pro Preview10.0
Opus 4.69.0
Opus 4.58.0
GPT-5.28.0
Qwen3.5 397B A17B4.0
GLM-52.0
Grok 41.0
Gemini 2.5 Pro0.0
Loading Atlas data…