Medmarks - Med-HALT Reasoning FCT: leaderboard
Metric: Score (%). Source: medmarks.ai. 71 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 94.56 |
| 2 | Grok 4 | 93.47 |
| 3 | GPT-5.1 (Medium) | 93.36 |
| 4 | Qwen 3 235B A22B (Thinking) | 90.07 |
| 5 | GPT-5.2 (Medium) | 89.98 |
| 6 | Claude Sonnet 4.5 | 89.73 |
| 7 | GPT-OSS-120B (Medium) | 87.96 |
| 8 | GPT-OSS-120B (High) | 87.67 |
| 9 | Qwen 3 Next 80B A3B (Thinking) | 86.86 |
| 10 | GPT-OSS-120B (Low) | 86.7 |
| 11 | GLM-4.7 FP8 | 86.17 |
| 12 | OLMo 3.1 32B (Thinking) | 81.91 |
| 13 | GPT-OSS-20B (High) | 80.53 |
| 14 | GPT-OSS-20B (Medium) | 75.74 |
| 15 | MiniMax-M2 | 75.6 |
Interactive version: theaggregate.ai/benchmark?slug=medmarks-med-halt-reasoning-fct · How It Works · Data refreshed daily, snapshot 2026-09-05.