Elorian releases Glimpse, its first visual reasoning model, as an API research preview

Glimpse's core mechanism is what Elorian calls a "visual scratchpad": most multimodal models convert an image into tokens once and then reason in text, whereas Glimpse can, mid-reasoning, crop to a specific region, extract a particular frame from a video, or run code to measure and compare what it sees. The model is released as a research preview with API access, and the chat interface is waitlist-only. Pricing is $0.60 per million input tokens and $3 per million output tokens, and the company says a small team trained it on a few hundred GPUs, a scale notably smaller than the frontier systems it benchmarks against. According to Elorian's evaluations, Glimpse surpasses GPT-6 Astra on scientific chart reasoning and short-video subtitles, with competitive results on other visual reasoning tasks. Its initial target customers are companies with large video libraries: robotics firms, scientific labs, and media and creative companies. Robotics teams, scientific labs, and media companies using the API can directly call this visual reasoning pipeline of cropping plus code execution; longer term, Elorian wants to move from "seeing" to "doing," using scene modeling to select actions, including directly controlling robots, and plans to build a model next year that leads on various visual reasoning tests.