Elorian’s first AI model wants to reason with images, not just about them
TechCrunch Aditya Mehta
Elorian just released Glimpse, an AI model built to look back at images and video while it reasons. It’s trying to make vision feel less like captioning a photo and more like actually seeing it.
Based on reporting by TechCrunch, Aditya Mehta — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Elorian has put its first model into the world, and it’s making a pretty specific bet: that most AI systems still treat images like a thing to describe, not a thing to inspect. The Palo Alto startup’s model, Glimpse, arrived Friday as a research preview with API access. The chat interface is still waitlist-only, which makes this less of a polished product launch than a public argument about where multimodal AI has been missing the point.
That argument is simple enough to grasp. Most multimodal models glance at an image, turn it into tokens, and then reason in text. Glimpse is meant to keep going back to the source while it thinks. It can crop into a region, pull specific frames from video, or run code to measure and compare what it sees. Elorian calls that a visual scratch space. Andrew Dai, the company’s CEO and a former Google DeepMind researcher, says the idea came into focus after he asked a frontier model who was winning in a board game photo and got nowhere useful.
The company is also trying to make the case that this can be built efficiently, not just by throwing giant compute budgets at the problem. Elorian says Glimpse was trained by a small team on a few hundred GPUs and is substantially smaller than the frontier systems it’s being compared with. Pricing sits at $0.60 per million input tokens and $3 per million output tokens, which puts it below models like Sonnet and Sol, but above lighter options like Haiku and Luna.
On benchmarks, Elorian says Glimpse beats even GPT-6 Astra on scientific chart reasoning and short-form video captioning, and posts competitive results elsewhere in visual reasoning. That matters because the company isn’t pitching this as a clever demo for image captions. The first customers it wants are the people drowning in video: robotics firms, scientific labs, and media and creative shops. The longer game is bigger still. Dai says Elorian wants to move from seeing to doing, eventually using scene models to choose actions and control robots.
My take — AI-written commentary, not fact-checked reporting
The useful part here is not the shiny benchmark talk, it’s the refusal to pretend that one-pass image tagging counts as vision. Too many AI companies still sell “multimodal” like it’s a new paint job on a text model. Building around inspection, not just description, is the right fight — and a small relief from the usual demo theater.
Read more about this at: TechCrunch