Elorian releases Glimpse, AI model that reasons with images

TL;DR: Elorian, a Palo Alto AI startup that raised $55 million before launching a product, has released Glimpse, its first AI model that can repeatedly examine images during reasoning rather than converting them to tokens once. The model outperforms GPT-6 Astra on scientific chart reasoning and short-form video captioning according to Elorian's evaluations, and is now available as a research preview with API access.

What's New

Elorian has released Glimpse, its first AI model designed to fundamentally change how artificial intelligence processes visual information. Unlike traditional multimodal models that look at an image once, convert it to tokens, and then reason about it in text, Glimpse can repeatedly return to the source image during its reasoning process.

The model uses what Elorian calls a "visual scratch space" — a mechanism that allows it to crop into specific regions, pull particular frames from videos, or run code to measure and compare what it's examining. CEO Andrew Dai, a former Google DeepMind researcher, compares it to how humans pinch to zoom on phone photos to double-check details.

Dai's inspiration came from a simple failure: asking a frontier AI model to determine who was winning in a photograph of a board game. The model couldn't answer. "Board games each have their own visual language, each have their own iconography, symbols," Dai explained. "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."

While the approach of crop-and-zoom during reasoning isn't entirely new — OpenAI added similar functionality to its o-series models in 2025 — Elorian claims to have built its entire model around this capability. The company says a small team trained Glimpse on just a few hundred GPUs, making it substantially smaller than the frontier systems it competes against.

According to Elorian's evaluations, Glimpse outperforms even GPT-6 Astra on scientific chart reasoning and short-form video captioning, with competitive results on other visual reasoning tasks.

Pricing and Availability

Glimpse is launching as a research preview with API access available now. A chat interface is also planned but currently operates on a waitlist-only basis.

The pricing structure is set at $0.60 per million input tokens and $3 per million output tokens. This positions Glimpse as cheaper than mid-level models like Sonnet and Sol, but more expensive than lightweight alternatives like Haiku and Luna.

Who It's For

Elorian is initially targeting companies that work extensively with video content. This includes robotics firms, scientific laboratories, media companies, and creative organizations that need advanced visual reasoning capabilities.

Longer term, the company has ambitious plans to move from visual understanding to action. Dai stated that Elorian wants to build a model of a scene and use that understanding to choose actions, including directly controlling robots.

Next year, the company aims to build "the world's best visual reasoning model" — one that leads on every visual reasoning test.

Company Background

Elorian emerged from stealth six months ago with a $55 million seed round at a $300 million valuation — raised before the company had launched any product. The round was co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean.

The Palo Alto-based startup's thesis centers on a fundamental limitation in current AI: that today's models, despite their skill with language and code, still can't truly see. As Dai put it: "Humans can look at Ikea diagrams, with no text, and assemble Ikea furniture, right? Why can't AI models?"