MMedBench: leaderboard
MMedBench — medical question answering across six languages; accuracy (%). Open-model rows use MMedBench full fine-tuning and have separate SFT identities; GPT and Gemini rows are zero-shot API evaluations.
Metric: Accuracy (%). Source: henrychur.github.io. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 74.27 | #438 |
| 2 | Gemini 1.0 Pro | 55.2 | #933 |
| 3 | GPT-3.5 | 51.82 | #818 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmedbench · How It Works · Data refreshed daily, snapshot 2026-10-20.