HealthBench Hard: leaderboard
Hard subset of HealthBench, evaluating difficult clinical and biomedical advice with physician-written rubrics and stricter scoring.
Metric: Score (%). Source: llm-stats.com. Status: saturated. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Muse Spark | 42.8 |
| 2 | GPT-5.6 Sol | 33.1 |
| 3 | GPT-5.6 Terra | 32.7 |
| 4 | GPT-5.6 Luna | 32 |
| 5 | GPT-OSS-120B | 30 |
| 6 | GPT-5.3 | 25.9 |
| 7 | GPT-5.5 Instant | 22.9 |
| 8 | GPT-OSS-20B | 10.8 |
| 9 | GPT-5 | 1.6 |
Interactive version: theaggregate.ai/benchmark?slug=healthbench-hard · How It Works · Data refreshed daily, snapshot 2026-09-05.