Elorian Launches Glimpse, an AI Model Built to Reason Directly With Visual Data
Glimpse can revisit images and video during reasoning, cropping regions, selecting frames, and running code to compare visual elements. Trained on a few hundred GPUs, it costs $0.60 per million input tokens and $3 per million output tokens.
Palo Alto AI startup Elorian released its first model, Glimpse, on Friday, October 17, 2026, in a research preview with API access. A chat interface is currently waitlist-only.
Glimpse is designed around a mechanism Elorian calls a “visual scratch space,” which allows the model to return to an image or video mid-reasoning — cropping into regions, pulling specific video frames, or running code to measure and compare visual elements. Most multimodal models, by contrast, convert an image to tokens once and then reason about it in text, effectively working from a description rather than the source material.
Andrew Dai, Elorian’s CEO and a former Google DeepMind researcher, said the limitation became clear to him during a board game session. He photographed a finished game and asked a frontier AI model who was winning. It couldn’t answer. “Board games each have their own visual language, each have their own iconography, symbols,” Dai told TechCrunch. “To an adult, it’s quite easy to understand what’s going on with minimal text, but these models aren’t able to intuit what the symbols mean.” He also pointed to Ikea assembly diagrams as an example of visual reasoning humans handle easily without text.
Elorian says Glimpse was trained by a small team on a few hundred GPUs and is substantially smaller than the frontier models it benchmarks against. Pricing is set at $0.60 per million input tokens and $3 per million output tokens. According to the company’s own evaluations, Glimpse outperforms GPT-6 Astra on scientific chart reasoning and short-form video captioning, with competitive results on other visual reasoning tasks.
Elorian came out of stealth six months ago after raising a $55 million seed round at a $300 million valuation, co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean.
The company is initially targeting robotics firms, scientific labs, media companies, and other organizations that work with large volumes of video. Longer term, Elorian aims to move from visual reasoning into action — including direct robot control. Dai said the company plans to build “the world’s best visual reasoning model” by next year, one that leads across every visual reasoning benchmark.