Elorian’s first AI model wants to reason with images, not just about them

Elorian released its first AI model called Glimpse, which approaches image understanding differently than existing models. Rather than looking at an image once and then reasoning about it in text, Glimpse repeatedly returns to the source image during reasoning, using techniques like cropping into regions or pulling specific video frames to verify what it observes. The company argues that current AI models cannot truly see, as demonstrated by their failure to understand board games or interpret visual symbols without text. According to Elorian's evaluations, Glimpse outperforms larger models on certain visual reasoning tasks while being substantially smaller and trained on fewer resources.