Elorian’s Glimpse looks again at the image while it reasons
Aditya Mehta, TechCrunch, 9:01 a.m. Pacific on Friday: Elorian, which raised a $55 million seed at a $300 million valuation before it had a product, released Glimpse as a research preview. The API is $0.60 per million input tokens and $3 per million output. Chief executive Andrew Dai, formerly of Google DeepMind, says the model crops, rewatches and measures instead of reasoning from a single look.
Short1 minEasy
Eli Navarro1 min read
- Elorian
- Glimpse
- Vision
- API

Photo: ElorianSource page
Elorian released Glimpse on Friday, a research preview with an API and a chat interface. Aditya Mehta at TechCrunch, in a story timestamped 9:01 a.m. Pacific, reported that the Palo Alto company had raised a $55 million seed round at a $300 million valuation before it had a product. The round was co-led by Striker Venture Partners, Menlo Ventures and Altimeter Capital, with Nvidia and Google chief scientist Jeff Dean participating. Andrew Dai, the chief executive and a former Google DeepMind researcher, told Mehta the model is built to return to an image or a video while it is still reasoning: crop a region, pull frames, or run code to measure what it is looking at. Elorian calls that a visual scratch space. The company’s blog, dated 16 October, prices the API at $0.60 per million input tokens and $3 per million output tokens.
Dai’s example for why one look is not enough was a board game. He photographed a finished game and asked a frontier model who was winning. It could not say. “Humans can look at Ikea diagrams, with no text, and assemble Ikea furniture, right? Why can’t AI models?” he told TechCrunch. Elorian says a small team trained Glimpse on a few hundred GPUs, post-training an open-weights base model, and that the system is substantially smaller than the frontier models in its table. On that table, which is the company’s, Glimpse is ahead of GPT-6 Astra on CharXiv-R, a scientific-chart test, 92.90 to 92.10, and on UGC-VideoCap, a short-form video-captioning test, 82.66 to 80.32. It matches Astra at 83.75 on ERQA, an embodied-reasoning score. The blog says both token rates are 94 percent lower than Astra’s. Those are Elorian’s prices and Elorian’s scores. Mehta wrote that the first customers Dai is aiming at are companies that already sit on video: robotics firms, scientific labs, and media and creative companies.


