Elorian launches Glimpse, an AI model built to actually see

Elorian, the Palo Alto AI startup that raised a $55 million seed round at a $300 million valuation before it had shipped a product, released its first model, Glimpse, on Friday, TechCrunch has exclusively learned. The model launches as a research preview with API access; a chat interface is waitlist-only, so for now the company has little more to show than a results table and a thesis: that today’s AI models, for all their skill with language and code, still can’t really see.

Most multimodal models look at an image once, convert it to tokens, and then reason about it in text — like describing a photo from memory. Glimpse instead keeps going back to the source: mid-reasoning, it can crop into a region, pull specific frames from a video, or run code to measure or compare what it’s looking at, a mechanism Elorian calls a “visual scratch space” — similar to pinching to zoom on a phone photo to double-check a detail. The approach isn’t entirely new — OpenAI added crop-and-zoom reasoning to its o-series models in 2025 — but Elorian says it built its whole model around the idea: a small team trained Glimpse on a few hundred GPUs, and the resulting system is substantially smaller than the frontier models it’s benchmarked against.

For Elorian CEO Andrew Dai, a former Google DeepMind researcher, the gap became obvious last year, when he photographed a board game he’d just finished with friends and asked a frontier model who was winning — it couldn’t say, because, he told TechCrunch, board games each have their own visual language and iconography that an adult can parse with minimal text but that the models can’t intuit. “Humans can look at Ikea diagrams, with no text, and assemble Ikea furniture, right? Why can’t AI models?” he said.

Glimpse costs $0.60 per million input tokens and $3 per million output tokens — more than lightweight models like Haiku and Luna, but less than mid-tier models like Sonnet and Sol. By Elorian’s own evaluations, Glimpse outperforms even GPT-6 Astra on scientific chart reasoning and short-form video captioning, with competitive results on other visual-reasoning tasks. The company came out of stealth six months ago with a round co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with Nvidia and Google chief scientist Jeff Dean also participating. Its initial targets are companies sitting on large volumes of video — robotics firms, scientific labs, media, and creative companies — and Dai said Elorian’s longer-term ambition is to move from seeing to doing, building a model of a scene that can choose an action, including directly controlling robots. Next year, he said, the company aims to build the best visual-reasoning model available, leading on every test in the category.

Why it matters

Glimpse is a bet that visual reasoning — not language fluency — is the next constraint on what AI models can actually do, especially for the video-heavy industries, like robotics and scientific research, it’s targeting first. If a small, cheaper model goes on to outperform frontier systems on visual tasks by building the capability in from the start, that’s a signal that narrow, purpose-built models could start chipping into big labs’ advantage in specific domains.