Elorian, a Palo Alto‑based artificial‑intelligence startup that secured a $55 million seed round at a $300 million valuation before unveiling any product, introduced its first model on Friday, TechCrunch reports exclusively. Named Glimpse, the system is available as a research preview with both API access and a chat interface.
The company’s premise is that current AI models, despite their prowess with language and code, still lack true visual comprehension. Most multimodal models process an image once, turn it into tokens, and then generate text‑based answers, essentially describing a photo from memory. Glimpse, by contrast, can revisit the original visual data during its reasoning. It can zoom into a region, extract specific video frames, or even execute code to measure or compare elements it sees, a capability Elorian calls a “visual scratch space.” The effect is similar to a person pinching a phone screen to double‑check a detail.
Former Google DeepMind researcher and Elorian CEO Andrew Dai pointed to a personal anecdote as evidence of the gap. Last year he photographed a board game he had just finished with friends and asked a leading model who was winning. The model could not answer. “Board games each have their own visual language, their own iconography, symbols,” Dai told TechCrunch. “To an adult it’s quite easy to understand what’s happening with minimal text, but these models aren’t able to intuit what the symbols mean. Humans can look at Ikea diagrams with no text and assemble the furniture, right? Why can’t AI models?”
The idea of dynamic cropping during reasoning is not entirely new—OpenAI added a similar feature to its o‑series models in 2025—but Elorian’s claim is that it built the entire architecture around it while keeping costs low. The team trained Glimpse on a few hundred GPUs, and the model is considerably smaller than the cutting‑edge systems it was benchmarked against. Pricing is set at $0.60 per million input tokens and $3 per million output tokens, making it cheaper than mid‑tier models such as Sonnet and Sol, though more expensive than lightweight offerings like Haiku and Luna. In Elorian’s internal tests, Glimpse outperformed even GPT‑6 Astra on scientific chart reasoning and short‑form video captioning, and delivered competitive results on other visual‑reasoning tasks.
Elorian emerged from stealth six months ago, raising a round co‑led by Striker Venture Partners, Menlo Ventures and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean. Its initial customers are expected to be companies that handle large volumes of video—robotics firms, scientific laboratories, media outlets and creative agencies. Looking farther ahead, the startup wants to move from merely seeing to acting, using visual models to select actions and eventually control robots directly. Dai said the goal for next year is to build “the world’s best visual reasoning model,” one that leads on every visual‑reasoning benchmark.
Aditya Mehta is an editorial fellow at TechCrunch covering AI. He is supported by the Tarbell Center for AI Journalism and attended UC Berkeley. He can be reached at the email address provided in the byline or via encrypted message on Signal.