Elorian launches Glimpse AI model with visual reasoning capabilities

Elorian, a Palo Alto AI startup, released its first model Glimpse on Friday. Unlike most multimodal models that convert images to tokens for text-based reasoning, Glimpse uses a 'visual scratch space' mechanism to crop regions, pull video frames, and run code during reasoning to measure or compare what it sees. The model, trained on a few hundred GPUs, outperforms GPT-4o Astra on scientific chart reasoning and short-form video captioning, with API pricing at $0.60 per million input tokens and $3 per million output tokens. Elorian aims to move from visual reasoning to direct action, including robot control.