JMed48k (With Images): leaderboard

JMed48k (2026) is built from the official question booklets of 11 Japanese national healthcare licensing examinations. This board is accuracy (%) of vision-language models on the 2,579 scored questions that come with images (clinical photographs, radiology, charts and others), the unweighted mean over the 11 professions; higher is better.

Metric: Accuracy (%) on the 2,579 scored JMed48k-Eval items that include images, unweighted macro average over the 11 professions, official MHLW gold answers, single-shot Japanese prompts at temperature 0 with reasoning disabled where the model has a switch, JSON-constrained answers; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 20 models tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 2.5 Pro79.7#276
2GPT-578.8#178
3GPT-5 Mini70.6#364
4Gemini 2.5 FlashNon-reasoning66.7#463
5Qwen 3.5 397B A17BNon-reasoning63.6#264
6Gemma 4 31B (IT)61.6#350
7Claude Sonnet 457.7#410
8Qwen 3.5 27BNon-reasoning57.6#375
9Llama 4 Maverick53.8#1110
10Grok 4.2053.6#301
11GLM-4.6V-FlashNon-reasoning52.8#1040
12Lingshu-32B45#1113
13Qwen 3.5 9BNon-reasoning41.8#828
14Hulu-Med-32B36.8#1221
15MedGemma-27B-IT34.2#1350

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=jmed48k-with-images · How It Works · Data refreshed daily, snapshot 2026-11-21.