HealthBench — Explicit no-tools; same range as bare HealthBench but kept separate (narrower model set).
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | GPT-5 | OpenAI | 67.2 | 1 | 2026-06-12 |
| 2 | Kimi K2 (Reasoning) | Moonshot | 58 | 2 | 2026-06-12 |
| 3 | DeepSeek-V3.2 | DeepSeek | 46.9 | 1 | 2026-06-12 |
| 4 | Claude Sonnet 4.5 | Anthropic | 44.2 | 1 | 2026-06-12 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.