How Good LLMs Are at Answering Bangla Medical Visual Questions? Dataset and Benchmarking

In three linesBanglaMedVQA: new benchmark for medical visual question answering in Bangla with clinically validated image-question-answer pairs. Evaluation of foundation models (Gemini, GPT-4.1 mini, Gemma-3) reveals substantially lower performance than English, severe limitations in fine-grained medical reasoning and specialized diagnostics.

Summary generated by Claude — human-verified