Elorian, a Palo Alto AI startup valued at $300 million after a $55 million seed round, has launched its first AI model named Glimpse. Released as a research preview on October 16, 2026, Glimpse offers an API and chat interface designed to advance how AI interacts with visual data. Unlike current multimodal models that analyze images once and then reason through text, Glimpse continually refers back to the image or video itself during its reasoning process. It can zoom in on sections, extract frames, and run code to better measure or compare visual elements, mimicking the way humans inspect and interpret images in detail.

The company’s approach addresses a key limitation in existing AI: inability to truly understand and reason about images beyond surface-level descriptions. Elorian CEO Andrew Dai, who previously worked at Google DeepMind, highlights examples like interpreting board games or Ikea diagrams, where human interpretation relies on visual symbols and context without textual cues. The model’s unique “visual scratch space” allows dynamic, iterative engagement with visuals, providing a richer understanding much like a human would have by revisiting and examining different parts of an image or scene.

Glimpse was developed efficiently by a small team using a few hundred GPUs and is smaller than many cutting-edge models, yet it demonstrates strong performance. According to Elorian's evaluations, Glimpse outperforms high-profile models such as GPT-6 Astra on tasks like scientific chart reasoning and short-form video captioning, while remaining competitive on other visual reasoning challenges. Pricing for Glimpse is set at $0.60 per million tokens in and $3 per million tokens out, positioning it as cost-effective compared to other mid-tier AI models.

Elorian emerged from stealth mode six months ago backed by investors including Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with Nvidia and Google chief scientist Jeff Dean participating. The startup aims initially to serve industries with extensive video data, such as robotics, scientific research, media, and creative sectors. Looking ahead, Elorian plans to move beyond visual reasoning toward actionable intelligence, enabling AI systems to build models of environments and directly control robots. By next year, the company targets creating the world’s leading model for visual reasoning across all benchmark tasks.