BanglaMedVQA (CoT) - MedICaT: leaderboard
The MedICaT subset of BanglaMedVQA (Ahmed et al., 2026): 1,000 short-answer questions about medical figures from the MedICaT image-caption corpus, asked with the Bangla questions and the chain-of-thought prompt. Exact-match accuracy in percent; higher is better.
Metric: Exact-match accuracy (%; 1,000 questions on MedICaT figures; Bangla questions, chain-of-thought prompt; exact match with the reference answer). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 3 12B | 40.5 | #2013 |
| 2 | Gemini 2.5 Flash | 34 | #452 |
| 3 | Qwen 2.5 VL 7B Instruct | 29 | #1842 |
| 4 | MedGemma-4B | 28 | #2460 |
| 5 | GPT-4.1 Mini | 16 | #775 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=banglamedvqa-cot-medicat · How It Works · Data refreshed daily, snapshot 2026-11-12.