Elorian, a Palo Alto AI startup, released its first model on Friday after raising $55 million at a $300 million valuation before launching any product, according to TechCrunch. The model, called Glimpse, is now available as a research preview through both API access and a chat interface. The company's argument is that current AI systems can't truly see — they glance at an image once, turn it into tokens, and then work from memory rather than continuing to look at what's in front of them.
Glimpse operates differently by returning to the original image mid-reasoning, allowing it to zoom into specific areas, extract individual frames from video, or execute code to measure and compare visual elements through what Elorian calls a "visual scratch space." The model costs $0.60 per million input tokens and $3 per million output tokens, placing it below mid-tier systems like Sonnet and Sol but above lightweight options like Haiku and Luna. Elorian says a small team built Glimpse using just a few hundred GPUs, making it substantially smaller than the frontier models it competes against. In the company's testing, Glimpse beats even GPT-6 Astra on scientific chart reasoning and short-form video captioning, while delivering competitive performance on other visual reasoning tasks.
Elorian CEO Andrew Dai, a former Google DeepMind researcher, told TechCrunch that the limits of AI vision became clear during a game night last year when he photographed a board game and asked a frontier model to identify the winner — it couldn't. "Board games each have their own visual language, each have their own iconography, symbols," Dai explained. He pointed to a simpler test: "Humans can look at Ikea diagrams, with no text, and assemble Ikea furniture, right? Why can't AI models?" The approach of cropping and zooming during reasoning isn't entirely new — OpenAI added it to its o-series models in 2025 — but Elorian built its entire system around this capability from the ground up, and claims to have done so cheaply.
The company emerged from stealth six months ago with funding co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with backing from Nvidia and Google chief scientist Jeff Dean. Its initial focus is on organizations handling large volumes of video: robotics companies, scientific laboratories, media operations, and creative firms. Looking ahead, Elorian plans to move beyond visual understanding to action, building models that can construct a representation of a scene and use it to make decisions, including direct robot control. Next year, Dai said, the company aims to build "the world's best visual reasoning model," one that leads on every visual reasoning test. For Glimpse to succeed beyond benchmarks, enterprise customers will need to trust a startup barely out of stealth with production workloads, and the model's visual scratch space will need to prove faster and more accurate than simply prompting larger frontier systems multiple times. The real test isn't whether Glimpse can see better than today's models, but whether seeing better solves problems valuable enough to justify switching providers.