Elorian launches Glimpse, an AI model that reasons with images

Elorian, an AI startup based in Palo Alto, released its first model, Glimpse, on Friday, October 16, 2026. The model is available as a research preview through an API, while its chat interface is open only to people on a waitlist.

The company argues that current AI models, however capable with text and code, still struggle to genuinely see. Most multimodal systems process an image once, turn it into tokens and then reason in text alone. Glimpse is designed to return to the original visual material repeatedly as it works through a problem.

A visual scratch space

Elorian calls the mechanism a "visual scratch space." During reasoning, Glimpse can crop and zoom into part of an image, extract chosen frames from a video, or execute code to measure and compare what it sees. The idea has precedent: OpenAI gave its o-series models crop-and-zoom abilities during reasoning in 2025. Elorian says its distinction is building the entire model around the technique, and doing so cheaply.

Chief executive Andrew Dai, a former Google DeepMind researcher, traces the project partly to a board game night last year, when a frontier model could not tell him who was winning from a photo of the table. He says games rely on their own symbols that adults grasp easily but models fail to interpret. "Humans can look at Ikea diagrams, with no text, and assemble Ikea furniture," Dai said.

According to Elorian, a small team trained Glimpse on a few hundred GPUs, and the model is considerably smaller than the frontier systems it is compared with. The company's own evaluations show it beating GPT-6 Astra on scientific chart reasoning and short-form video captioning, and performing competitively on other visual reasoning benchmarks. These results have not yet been independently verified.

Pricing, backers and ambitions

Glimpse is priced at $0.60 per million input tokens and $3 per million output tokens, below mid-tier models such as Sonnet and Sol but above lightweight options like Haiku and Luna. Elorian emerged from stealth six months ago after raising a $55 million seed round at a $300 million valuation before shipping any product. Striker Venture Partners, Menlo Ventures and Altimeter Capital co-led the round, with Nvidia and Google chief scientist Jeff Dean also participating.

Early customers are expected to be organisations with large video archives, including robotics companies, scientific labs, media outlets and creative firms. Over time, Elorian wants its models to move from perception to action, building a representation of a scene and using it to choose actions, including controlling robots directly. Dai said the goal for next year is a model that leads every visual reasoning benchmark.