MBABench 2.0 — Medium-Hard
Professional Work · 2026-10-10
Strict numerical and workbook-quality pass rate on the thirty Medium-Hard tasks.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 3.3 |
| Claude Opus 5 | 3.3 |
| GPT-6 Astra | 3.3 |
| Grok 4.6 | 0.0 |
| Kimi K3 | 0.0 |