MBABench 2.0 — Easy Numerical Accuracy
Professional Work · 2026-10-10
All-answers-correct rate on the eleven Easy tasks, without the workbook-quality gate.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 90.9 |
| GPT-6 Astra | 90.9 |
| Grok 4.6 | 90.9 |
| Claude Opus 5 | 72.7 |
| Kimi K3 | 54.5 |