Elorian releases Glimpse visual reasoning model in research preview
Elorian debuts Glimpse with API access
Elorian, the Palo Alto AI startup that raised a $55 million seed round at a $300 million valuation before launching a product, has released its first model, Glimpse. The model is available as a research preview through an API, while its chat interface is limited to a waitlist.
Glimpse is designed for visual reasoning. Elorian argues that many multimodal systems examine an image once, translate it into tokens and then continue reasoning primarily in text. Its model instead returns to the underlying visual material during the reasoning process.
That approach is implemented through what Elorian calls a visual scratch space. The model can crop into a region of an image, retrieve selected frames from a video, or run code to measure and compare elements it is examining. The intended effect is closer to checking a detail again than relying on an initial description of a scene.
A model built around iterative visual inspection
Andrew Dai, Elorian’s chief executive and a former Google DeepMind researcher, has pointed to board games and text-free IKEA instructions as examples of the gap. Such material depends on visual conventions, icons and symbols that people can often interpret with little or no written explanation.
The concept has precedents: OpenAI added crop-and-zoom capabilities during reasoning to its o-series models in 2025. Elorian’s claim is that Glimpse was built around this capability from the outset and trained comparatively efficiently. The company says a small team used a few hundred GPUs, and that Glimpse is substantially smaller than the frontier models used as benchmarks.
Elorian’s public release follows the funding described in Elorian's $55 million seed funding and positions the company around visual reasoning rather than a general-purpose chatbot. Its initial customers are expected to include robotics firms, scientific laboratories, media organizations and creative businesses holding large volumes of video.
Benchmarks, pricing and the route to action
In Elorian’s own evaluations, Glimpse surpassed GPT-6 Astra on scientific chart reasoning and short-form video captioning, while producing competitive results on other visual-reasoning tasks. These are company-reported results, and the research preview gives prospective users a way to assess performance on their own visual material.
Glimpse is priced at $0.60 per million input tokens and $3 per million output tokens. Elorian places those rates below mid-level models such as Sonnet and Sol, but above lightweight models including Haiku and Luna.
Over the longer term, Elorian intends to use scene understanding to select actions, including direct robot control. For businesses, the immediate implication is to evaluate iterative inspection on bounded tasks involving charts, video or technical images, where the ability to revisit evidence may matter more than a single-pass visual summary.