MBABench 2.0
Professional Work · 2026-10-10
101 financial-modeling spreadsheet tasks authored by finance experts across six finance areas and sixteen model types. A task passes only when all numerical answers are correct and the workbook meets the professional quality thresholds. Harness and effort are reported separately for each agent.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 15.8 |
| Claude Fable 5.1 | 7.9 |
| Claude Opus 5 | 4.0 |
| Grok 4.6 | 1.0 |
| Kimi K3 | 0.0 |