According to an exclusive report from TechCrunch, Elorian, an AI startup based in Palo Alto, has raised $55 million in seed funding at a $300 million valuation before officially launching its product. The company released its first model on Friday.
The model, named Glimpse, is being released as a research preview with API access; the chat interface is currently available only to users on the waitlist. For now, it functions more like a results table and a research paper. Its core claim is that today’s AI models, while strong in language and code, still “don’t see clearly.”
Most multimodal models first glance at an image, convert it into tokens, and then perform reasoning with text—like describing a photo from memory. Glimpse, however, continually returns to the source image. During reasoning, it can zoom into a specific region, extract particular frames from a video, or run code to measure or compare what it is viewing. Elorian refers to this mechanism as a “visual scratch space.” Think of it as someone pinching and zooming in on a photo on their phone, repeatedly verifying what they just saw.
For Andrew Dai, former Google DeepMind researcher and CEO of Elorian, one of the most obvious signs that AI still cannot truly “see” emerged during a game night.
Last year, he took a photo of a board game he had just finished playing with friends and asked a cutting-edge model who was leading. The model could not answer. He told TechCrunch that board games “each have their own visual language, icon systems, and symbols.” “For adults, very little text is needed to understand what’s happening, but these models cannot intuitively grasp the meaning of these symbols.”
Humans can look at IKEA’s instruction diagrams—without any text—and still assemble IKEA furniture, right? Why can’t AI models do the same?” he said.
This approach isn’t entirely new—OpenAI introduced truncation and scaling during inference for its o-series models in 2025—but Elorian claims it has built the entire model around this concept at a lower cost. The company says only a small team trained Glimpse using hundreds of GPUs, and the model is significantly smaller than comparable state-of-the-art systems.
Glimpse is priced at $0.60 per million input tokens and $3 per million output tokens. It is cheaper than mid-tier models like Sonnet and Sol, but more expensive than lightweight models like Haiku and Luna.
However, according to Elorian’s evaluation, Glimpse outperforms GPT-6 Astra in scientific chart reasoning and short video subtitle generation, and is also competitive in other visual reasoning tasks.
Elorian emerged from stealth six months ago, following a funding round co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with participation from Nvidia and Google Chief Scientist Jeff Dean.
The company initially targeted enterprises with large volumes of video content: robotics companies, research laboratories, and media and creative firms. In the long term, Elorian aims to move from “seeing” to “acting”—this means building a scene model and using it to select actions, including directly controlling robots. Dai said the company plans to build “the world’s best visual reasoning model” next year, leading in all visual reasoning benchmarks.