Atlas

Benchmarks

← All benchmarks

SLAKE

Health & Medicine · 2021-02-18

SLAKE is a bilingual medical visual question-answering dataset of radiology images (CT, MRI and X-ray) with semantic labels, a medical knowledge graph and open- and closed-ended questions. Release tables report answer accuracy and do not always identify the language subset.

Top models (higher is better)

ModelScore
Qwen3.8-Max90.8
GPT-5.6 Sol85.1
Qwen3.5-Omni-Plus84.7
Qwen3.7-Plus (source unspecified)83.2
Gemini 3.1 Pro Preview82.9
Qwen3.5 Plus (2026-02-15)82.8
Kimi K2.581.6
Qwen3.5 122B A10B81.6
Gemini 3 Pro Preview81.3
Qwen3.5-27B80.0
Qwen3.5 397B A17B79.9
Qwen3.5-9B79.0
Qwen3.5 35B A3B78.7
GPT-5.276.9
Opus 4.576.4
Qwen3.5-4B76.1
Opus 4.875.9
Qwen3.5-2B74.4
Sonnet 4.573.6
Qwen3.5-Omni-Flash73.1
GPT-5 Mini70.5
Gemini 2.5 Flash-Lite65.0
Qwen3.5-0.8B62.6
GPT-5 Nano57.0
Loading Atlas data…