HealthBench Hard — leaderboard
Hard subset of HealthBench, evaluating difficult clinical and biomedical advice with physician-written rubrics and stricter scoring.
Metric: Overall score (self-reported). Source: benchmarklist.com. Status: saturation imminent. 55 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-OSS-120B | 60 |
| 2 | Gemma 3 27B | 59 |
| 3 | Qwen 3 30B A3B 2507 (Thinking) | 58 |
| 4 | Qwen 3 8B | 56 |
| 5 | Qwen 3 4B 2507 (Thinking) | 54 |
| 6 | Qwen 3 32B | 50 |
| 7 | Qwen 3 235B A22B | 50 |
| 8 | Qwen 2.5 72B Instruct | 49 |
| 9 | GPT-OSS-20B | 48 |
| 10 | DeepSeek R1 Distill Llama 70B | 47 |
| 11 | Qwen 3 4B | 43 |
| 12 | Qwen 2.5 7B Instruct | 42 |
| 13 | DeepSeek V3 | 42 |
| 14 | GPT-5 | 41.6 |
| 15 | GPT-5.1 | 41.4 |
Interactive version: theaggregate.ai/benchmark?slug=healthbench-hard · How the rankings work · Data refreshed daily, snapshot 2026-07-22.