MedQ-Deg - Treatment Reasoning: leaderboard
Metric: Accuracy (%) on the Treatment questions of MedQ-Deg (capability dimension of the clinical reasoning branch), pooled over the L1 (mild) and L2 (severe) degraded images, multiple-choice medical VQA items merged from OmniMedVQA, GMAI-MMBench and MedXpertQA, each image corrupted by modality-specific degradations whose severity three radiologists calibrated (24,894 QA pairs over 7 modalities and 18 degradation types), single-letter answer, temperature 1.0, mean of 3 runs; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 40 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 78.8 | #91 |
| 2 | Gemini 2.5 Pro | 59.6 | #145 |
| 3 | Claude Sonnet 4.5 | 53.9 | #138 |
| 4 | GPT-4o | 46.1 | #333 |
| 5 | Gemini 2.5 Flash | 46.1 | #237 |
| 6 | GPT-4.1 | 42.3 | #240 |
| 7 | GPT-4.1 Mini | 40.4 | #346 |
| 8 | GPT-5.1 | 38.5 | #131 |
| 9 | Gemma 3 4B | 26.9 | #1084 |
| 10 | MedGemma-4B | 25 | #731 |
| 11 | Gemma 3 27B | 19.2 | #596 |
| 12 | Mistral Small 3.1 | 19.2 | #600 |
| 13 | Qwen 3 VL 30B A3B Instruct | 19.2 | #365 |
| 14 | GPT-4o Mini | 15.4 | #588 |
| 15 | Qwen 3 VL 235B A22B Instruct | 15.4 | #264 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=medq-deg-treatment-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-11.