BankerToolBench Pass@3
Professional Work · 2026-04-13
Share of BankerToolBench v1.0's 100 tasks on which the best of three runs scores above the 0.80 rubric-score acceptance threshold. It is a component of the v1.0 factor group.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.4 | 23.0 |
| GPT-5.2 | 18.0 |
| Gemini 3.1 Pro Preview | 16.0 |
| Opus 4.6 | 12.0 |
| GLM-5 | 6.0 |
| Qwen3.5 397B A17B | 6.0 |
| Grok 4 | 3.0 |
| Gemini 2.5 Pro | 0.0 |