Elorian Launches Glimpse, an AI Model That Reasons with Images

Elorian, a Palo Alto startup that raised a $55 million seed round, released its first AI model Glimpse. The model is a research preview with API access, and a chat interface is on a waitlist only. Glimpse is built to show that current AI systems still cannot truly see. Most multimodal models read an image once, turn it into text tokens, and then reason about it.

Visual Scratch Space

Glimpse uses a visual scratch space, letting the model crop, zoom, pull frames from a video, or run code to measure what it sees. This works like a person pinching a photo on a phone to double‑check details. The model was trained on a few hundred GPUs, making it smaller than many frontier systems. Its pricing is $0.60 per million tokens in and $3 per million tokens out, cheaper than mid‑level models but more expensive than lightweight ones.

Elorian left stealth six months ago, raising a round led by Striker Venture Partners, Menlo Ventures and Altimeter Capital, with Nvidia and Google chief scientist Jeff Dean also invested. The company targets firms that work with large video collections, such as robotics, scientific labs, media and creative studios. In the long term, Elorian wants Glimpse to move from just seeing to deciding what to do, including controlling robots. Founder Andrew Dai said the team will build the world’s best visual‑reasoning model next year.