Elorian’s first AI model aims to reason using images rather than merely reason about them

Meta description: Elorian launched Glimpse, a research-preview visual reasoning model with crop-and-zoom, API access, $0.60 and $3 per million token pricing, and robotics goals.

Tags: Elorian, Glimpse, visual reasoning, AI models, multimodal

Featured image: Abstract technology style visual showing interconnected frames, crops, and measurement overlays; no real people, no company logos, no text.

Background: A Funding Round Ahead of a Product and a Perceptual Gap

Elorian's Glimpse visual reasoning model moved from a stealth startup story to a public research preview on Friday, TechCrunch exclusively learned. The Palo Alto AI startup had raised a $55 million seed round at a $300 million valuation before launching a product, and the model is meant to address a gap the company says many current multimodal systems still have: they process language and code well, but they do not reliably see.

Six months ago, Elorian came out of stealth with a seed round co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean. The company had not released a model yet. Its stated thesis is that most multimodal models look at an image once, convert it into tokens, and then reason in text, similar to describing a photo from memory.

Andrew Dai, a former Google DeepMind researcher and Elorian's CEO, described the problem through a board game example. Last year, after a game night, he photographed a finished board game and asked a frontier model who was winning. It could not answer. Dai said board games have their own visual language, iconography, and symbols, and while an adult can often understand what is happening with little text, models cannot always infer what the symbols mean. He also compared the challenge to assembling Ikea furniture from an Ikea diagram with no text, a task humans can do by looking.

Development: Visual Scratch Space, a Small Team, and a Smaller Model

Glimpse launches as a research preview with API access and a chat interface. The core mechanism Elorian calls a visual scratch space lets the model return to the source during reasoning. Instead of relying only on the initial tokenized view, it can crop into a region, pull specific frames from a video, or run code to measure or compare what it is looking at. The company frames this as closer to a person zooming into a phone photo to check a detail.

Elorian acknowledges the idea is not entirely new: OpenAI added crop-and-zoom during reasoning to its o-series models in 2025. The startup's pitch is that it built the entire model around this process and did so cheaply. The company says a small team trained Glimpse on a few hundred GPUs, and that the model is substantially smaller than the frontier systems it is benchmarked against.

The company also gave pricing figures. Glimpse costs $0.60 per million tokens in and $3 per million tokens out. Elorian positions that as cheaper than mid-level models such as Sonnet and Sol, but more expensive than lightweight models such as Haiku and Luna.

Latest Progress: Benchmarks, Initial Customers, and the Road From Seeing to Acting

According to Elorian's evaluations, Glimpse outperforms even GPT-6 Astra on scientific chart reasoning and short-form video captioning. It also produces competitive results on other visual reasoning tasks. Those results are presented as the company's own assessments.

The first customers Elorian wants to serve are organizations that sit on large amounts of video: robotics firms, scientific labs, media companies, and creative companies. That focus matches the model's mechanism, because video, charts, and visual scenes require repeated inspection, comparison, and measurement rather than a single static description.

Longer term, Dai said Elorian wants to move from seeing to doing. That means building a model of a scene and using that model to choose an action, including directly controlling robots. The company's stated goal for next year is to build the world's best visual reasoning model, one that leads on every visual reasoning test.

Glimpse is currently a research preview, not a broad production system. The next phase for Elorian will be whether its visual scratch space approach can turn early benchmark claims into product traction among video-heavy customers, while the company works toward the more ambitious robotics and action-oriented visual reasoning goal.