Elorian launches Glimpse, an AI model that re-examines images while reasoning

Elorian, a Palo Alto AI startup founded by former Google DeepMind researcher Andrew Dai, launched its first model on October 16, 2026. Glimpse is available now as a research preview with API access; a chat interface remains waitlist-only.

Most multimodal AI models look at an image once, convert what they see into text-like tokens, and then reason entirely in that text space from there — meaning fine visual detail that didn't make it into the initial conversion is simply gone for the rest of the task. Glimpse instead returns to the original image repeatedly during reasoning: it can crop specific regions, pull individual frames from video, or run code to measure or compare what it's looking at. Elorian calls this a "visual scratch space." The underlying idea isn't entirely novel — OpenAI added crop-and-zoom capability to its o-series models in 2025 — but Elorian built Glimpse around the technique from the ground up rather than bolting it onto an existing architecture.

The model was trained by a small team on a few hundred GPUs and is substantially smaller than the frontier systems it's being compared against, which shows in its pricing: $0.60 per million input tokens and $3 per million output tokens, positioning it above lightweight models but below mid-tier ones. According to Elorian's own benchmarks, Glimpse outperforms GPT-6 Astra on scientific chart reasoning and short-form video captioning and is competitive on other visual reasoning tasks — figures that, as company-reported results, warrant independent verification before being taken as settled.

Elorian came out of stealth six months ago and raised a $55 million seed round at a $300 million valuation before shipping a product, co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean — a notable show of confidence from figures with direct visibility into frontier AI research. Dai has argued that current models, despite strong language and coding ability, still struggle with tasks humans find trivial, like reading symbols on a board game or assembling furniture from a diagram without accompanying text.

Elorian is initially targeting robotics firms, scientific labs, and media companies that work with large volumes of video, and says its longer-term ambition is to move from passive perception to controlling physical systems like robots, with a goal of building what it calls the best visual reasoning model in the field within a year, as first reported by TechCrunch.