Elorian Unveils Glimpse, a Visual Reasoning Model That Re‑examines Images While Thinking

Elorian, a Palo Alto‑based AI startup, has released its first model called Glimpse as a research preview. The model introduces a “visual scratch space” that lets the system revisit an image during reasoning, cropping regions, pulling video frames, or running lightweight code to measure what it sees.

How Glimpse Differs from Typical Multimodal Models

Most multimodal models convert an image to tokens once and then reason about it in text, akin to describing a photo from memory. Glimpse, by contrast, can loop back to the raw pixels multiple times, allowing it to verify hypotheses much like a user pinching to zoom on a smartphone picture.

The company says Glimpse was trained on a few hundred GPUs and is considerably smaller than frontier systems. Priced at $0.60 per million input tokens and $3 per million output tokens, it sits between lightweight and mid‑range offerings. Early benchmarks show it outperforming GPT‑6 Astra on scientific chart reasoning and short‑form video captioning while remaining competitive on other visual tasks.

Elorian’s initial customers are expected to be robotics firms, scientific labs, and media companies that handle large video archives. Long‑term, the startup aims to extend visual reasoning into action, using scene models to drive robotic control.