Visual MCQ

Researchers found that test-time scaling improves small vision-language models on multilingual visual multiple-choice questions, but the largest gains come from fixing prompt parseability and increasi…