Medmarks - MedHallu Hard — leaderboard
Metric: Score (%). Source: medmarks.ai. 71 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 74.51 |
| 2 | GPT-5.1 (Medium) | 60.85 |
| 3 | Claude Sonnet 4.5 | 59.96 |
| 4 | Ling-flash-2.0 | 54.44 |
| 5 | Hermes 4 70B | 52 |
| 6 | GLM-4.7 FP8 | 51.99 |
| 7 | Qwen 3 235B A22B (Thinking) | 51.72 |
| 8 | Granite 4.0 H Small | 50.86 |
| 9 | Hermes-4-14B | 50.76 |
| 10 | Llama 3.1 8B Instruct | 49.42 |
| 11 | GPT-OSS-120B (High) | 48.88 |
| 12 | Phi-4 (Reasoning) | 48.36 |
| 13 | Qwen 3 4B (Reasoning) | 47.93 |
| 14 | GPT-OSS-20B (Low) | 47.28 |
| 15 | Llama 3.3 70B Instruct | 47.12 |
Interactive version: theaggregate.ai/benchmark?slug=medmarks-medhallu-hard · How the rankings work · Data refreshed daily, snapshot 2026-07-22.