Elorian, a Palo Alto AI startup that raised $55 million in a seed round at a $300 million valuation before shipping a product, released its first model on Friday, TechCrunch has learned.
Called Glimpse, the model is available as a research preview with API access; a chat interface is currently waitlist-only. Elorian frames the release around a central thesis: while modern AI models are adept at language and code, they still struggle to truly "see."
Most multimodal systems inspect an image once, convert it into tokens, and then reason about those tokens in text—similar to describing a photo from memory. Glimpse differs by repeatedly returning to the image source during the reasoning process. Mid-reasoning it can crop into a specific region, pull particular frames from a video, or run code to measure or compare visual content. Elorian refers to this mechanism as a "visual scratch space," likening it to someone pinching to zoom on a phone photo to double-check a detail.
Andrew Dai, Elorian’s CEO and a former Google DeepMind researcher, shared an anecdote about the limitation: last year he photographed a board game played with friends and asked a frontier model who was winning; the model couldn’t answer. "Board games each have their own visual language, their own iconography, symbols," Dai told TechCrunch. "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." He added that humans can follow IKEA diagrams without text, and questioned why AI systems cannot perform similarly.
The idea of cropping and zooming during reasoning is not entirely novel—OpenAI added a crop-and-zoom capability to its o-series models in 2025—but Elorian argues it built the entire model around that approach and did so inexpensively. The company says a small team trained Glimpse on a few hundred GPUs, and that the model is substantially smaller than the frontier systems it is benchmarked against.
In pricing, Glimpse costs $0.60 per million input tokens and $3 per million output tokens. Elorian positions that as cheaper than mid-tier models such as Sonnet and Sol, while pricier than lightweight models like Haiku and Luna.
According to Elorian’s internal evaluations, Glimpse outperforms even GPT-6 Astra on scientific chart reasoning and short-form video captioning, and delivers competitive results on other visual reasoning benchmarks.
Elorian emerged from stealth six months ago after closing a round co-led by Striker Venture Partners, Menlo Ventures and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean.
Initial target customers are companies with large video inventories—robotics firms, scientific laboratories, and media and creative businesses. Over the longer term, Elorian aims to move from "seeing to doing": building a model of a scene and using that representation to select actions, including direct robot control. Dai said the company aims next year to build "the world’s best visual reasoning model," one that leads on every visual reasoning test.
