Medmarks - MedHallu Hard: leaderboard
The "MedHallu Hard" subtask of Medmarks. Living medical-model benchmark spanning verifiable and open-ended clinical evaluation families with model/system comparisons.
Metric: Score (%). Source: medmarks.ai. Saturation forecast: Around 2029. 89 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 74.51 | #66 |
| 2 | Gemma 4 31B (IT) | 64.22 | #176 |
| 3 | Qwen 3.6 27B (Thinking) | 62.99 | #170 (Qwen 3.6 27B) |
| 4 | Qwen 3.6 35B A3B (Thinking) | 62.26 | #226 (Qwen 3.6 35B A3B) |
| 5 | GPT-5.1 (Medium) | 60.85 | #135 (GPT-5.1) |
| 6 | Claude Sonnet 4.5 | 59.96 | #142 |
| 7 | Ling-flash-2.0 | 54.44 | #550 |
| 8 | Qwen 3.5 27B (Thinking) | 53.07 | #187 (Qwen 3.5 27B) |
| 9 | Qwen 3.5 35B A3B (Thinking) | 52.96 | #256 (Qwen 3.5 35B A3B) |
| 10 | Hermes 4 70B | 52 | #506 |
| 11 | GLM-4.7 FP8 | 51.99 | #209 |
| 12 | Qwen 3 235B A22B (Thinking) | 51.72 | #313 (Qwen 3 235B A22B) |
| 13 | Granite 4.0 H Small | 50.86 | #665 |
| 14 | Hermes-4-14B | 50.76 | #751 |
| 15 | MedGemma-4B-IT | 50.37 | #853 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=medmarks-medhallu-hard · How It Works · Data refreshed daily, snapshot 2026-10-20.