Medmarks - Med-HALT Reasoning NOTA: leaderboard
Metric: Score (%). Source: medmarks.ai. 71 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 Next 80B A3B (Thinking) | 78.86 |
| 2 | Grok 4 | 78.54 |
| 3 | Gemini 3 Pro (Preview) | 74.4 |
| 4 | OLMo 3.1 32B (Thinking) | 72.07 |
| 5 | GPT-5.1 (Medium) | 69.85 |
| 6 | GPT-5.2 (Medium) | 69.75 |
| 7 | GLM-4.7 FP8 | 69 |
| 8 | Qwen 2.5 32B Instruct | 68.78 |
| 9 | Qwen 3 235B A22B (Thinking) | 68.29 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 66.97 |
| 11 | OLMo 3 32B (Thinking) | 63.53 |
| 12 | Ling-flash-2.0 | 61.51 |
| 13 | MiniMax-M2.1 | 61.08 |
| 14 | GPT-OSS-120B (Medium) | 60.56 |
| 15 | MiniMax-M2 | 59.86 |
Interactive version: theaggregate.ai/benchmark?slug=medmarks-med-halt-reasoning-nota · How It Works · Data refreshed daily, snapshot 2026-09-05.