OfficeQA — Anthropic office-productivity-document QA eval, base/non-Pro variant.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | Anthropic | 86.3 | 2 | 2026-06-21 |
| 2 | Claude Mythos 5 | Anthropic | 79 | 1 | 2026-07-24 |
| 3 | Opus 5 | Anthropic | 78.1 | 1 | 2026-07-24 |
| 4 | Claude Opus 4.8 | Anthropic | 77.6 | 1 | 2026-07-24 |
| 5 | Claude Opus 4.6 | Anthropic | 73.5 | 2 | 2026-06-21 |
| 6 | Claude Sonnet 5 | Anthropic | 73.3 | 1 | 2026-07-07 |
| 7 | Claude Sonnet 4.6 | Anthropic | 68.7 | 1 | 2026-07-07 |
| 8 | GPT-5.4 | OpenAI | 68.1 | 2 | 2026-06-21 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.