Elorian releases first AI model, Glimpse, which analyzes images by repeatedly returning to the source
Palo Alto startup Elorian, which raised a $55 million seed round, on Friday released its first model, Glimpse. During image and video analysis it can zoom into a frame or run code instead of relying on a single text description.
Elorian, an artificial intelligence startup based in Palo Alto, California, released its first model, Glimpse, on Friday, according to an exclusive report by technology outlet TechCrunch. Before shipping any product, the company had raised a $55 million seed round at a $300 million valuation. The model is currently a research preview: developers can reach it through an API (a programming interface), while the chat interface is waitlist-only. For now, it can be judged only by a results table and the company's central thesis.
What the model does differently
Elorian argues that today's models, despite their skill with language and code, still cannot really see. Most multimodal models, which handle both text and images, look at an image once, convert it to tokens and then reason in text, much like describing a photo from memory. Glimpse goes back to the original while it reasons: it can crop a region, pull specific video frames or run code to measure or compare what it sees. The company calls this mechanism a "visual scratch space."
Elorian CEO Andrew Dai, a former Google DeepMind researcher, says he noticed the problem after a game night. Last year he photographed a board game he had just finished with friends and asked a frontier model who was winning, but it could not answer. In his view, people easily grasp a game's symbols, while the models cannot work out what they mean.
The approach is not entirely new: OpenAI added crop-and-zoom during reasoning to its o-series models in 2025. Elorian's distinction is that it built the whole model around the idea, and did so cheaply. The company says a small team trained Glimpse on a few hundred GPUs and that the model is substantially smaller than the frontier systems it was benchmarked against.
Pricing and results
Glimpse costs $0.60 per million input tokens and $3 per million output tokens. That is cheaper than mid-level models such as Sonnet and Sol, but more expensive than lightweight models such as Haiku and Luna. According to Elorian's own evaluations, the model beats even GPT-6 Astra at scientific chart reasoning and short-form video captioning, with competitive results on other visual reasoning tasks.
Plans ahead
Elorian emerged from stealth six months ago with a round co-led by Striker Venture Partners, Menlo Ventures and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean. Its first targets are companies with large amounts of video: robotics firms, scientific labs, and media and creative companies. Longer term, it wants to move from seeing to acting by building a model of a scene, using it to choose an action and directly controlling robots. Next year, Dai said, the company aims to build "the world's best visual reasoning model."
