SurgCoT: leaderboard

Spatiotemporal reasoning on surgical videos of SurgCoT (2026), multiple-choice questions over 2,841 videos from seven specialties, answered zero-shot with only the video and the main question, averaged over five tasks (causal action ordering, cue-action alignment, affordance mapping, micro-transition localization, anomaly onset tracking). Main-question accuracy in percent; higher is better.

Metric: Average over the five reasoning tasks of main-question accuracy (%) on SurgCoT surgical-video multiple-choice questions (2,841 videos across 7 specialties and 35 procedures), baseline setting with only the video and the main question, zero-shot, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-576.62#178
2Qwen 3 VL 8B75.44#917
3Claude Sonnet 4.574.1#261
4MedGemma-27B-IT70.96#1350
5HuatuoGPT-Vision-7B70.08#2162
6Gemini 2.5 Pro70.02#276
7Qwen 2.5 VL 7B68.85#1862
8LLaVA-Med-7B68.15#6296
9InternVL3.5-8B67.95#1604
10Lingshu-7B63.43#1635

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

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