SAB Research Strict Pass Rate
Finance & Accounting
Share of SAB's 960 research tasks a model passes under the strict rule: a schema-valid answer with at least 80 points including 10 for evidence, every critical criterion correct and no nonexistent citation. It comes from the same runs and v0.7.1 answer keys as the research score; the release designates it as its primary leaderboard metric.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 95.7 |
| Claude Fable 5.1 | 94.2 |
| Claude Opus 5 | 93.6 |
| Grok 4.6 | 90.4 |
| GPT-6 Sol | 89.8 |
| GPT-6 Astra | 88.1 |
| GPT-5.6 Sol | 87.0 |
| GPT-5.6 Terra | 84.6 |
| GPT-5.6 Luna | 82.4 |
| GPT-6 Luna | 76.3 |
| Sonnet 5 | 75.4 |
| DeepSeek-V3.2 | 75.1 |
| Haiku 4.5 | 67.0 |