Colon-Bench - VQA (Unprompted): leaderboard
Metric: Accuracy (%) on Colon-Bench's 2,740 unprompted five-option video questions over 918 raw colonoscopy clips, lesion-free windows included; unanswered questions count as wrong; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 82.5 | #77 |
| 2 | Gemini 3 Flash | 76 | #93 |
| 3 | Seed-1.6 | 72 | #257 |
| 4 | GPT-5.2 | 71.4 | #105 |
| 5 | GPT-5.4 | 70.8 | #76 |
| 6 | Gemini 3.1 Flash Lite | 67.7 | #174 |
| 7 | Qwen 3.5 Plus | 60.5 | #123 |
| 8 | Qwen 3.5 397B A17B | 60.1 | #141 |
| 9 | Qwen 3 VL 235B A22B | 56.6 | #228 |
| 10 | GLM-4.6V | 53.8 | #309 |
| 11 | Molmo2-8B | 53.4 | #522 |
| 12 | Gemini 2.5 Flash Lite | 46.8 | #413 |
| 13 | Seed 1.6 Flash | 45.4 | #427 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=colon-bench-vqa-unprompted · How It Works · Data refreshed daily, snapshot 2026-10-11.