Test-Time Scaling for Small VLMs on Multilingual Visual MCQ
Why giving small AI models more thinking space beats fancier reasoning tricks
Small vision-language models can match larger ones on multilingual image questions through test-time scaling, but not in the way researchers expected. The real bottleneck isn't choosing between different reasoning paths—it's simply giving the model enough tokens to finish writing its answer. Fixing how the model formats its response and doubling the thinking budget from 1,000 to 2,000 tokens recovered most gains, while elaborate verification methods added almost nothing.
Smaller AI models are cheaper to run and deploy on phones or edge devices, but they've lagged behind large models on complex reasoning tasks. This work shows that smaller models don't need expensive new techniques—they just need simpler fixes and room to think. The approach achieved top performance on a major visual reasoning benchmark, suggesting that efficiency and capability aren't as far apart as once thought.