Elorian releases Glimpse, a visual reasoning model with API access

Elorian CEO Andrew Dai, a former Google DeepMind researcher, argues that current AI lacks true visual intuition, citing an example where a frontier model failed to identify the winner of a board game from a photo. Elorian claims Glimpse was trained by a small team on a few hundred GPUs, making it substantially smaller than frontier systems. The pricing is set at $0.60 per million input tokens and $3 per million output tokens. According to Elorian’s internal evaluations, Glimpse outperforms GPT-6 Astra on scientific chart reasoning and short-form video captioning, though it is more expensive than lightweight models like Haiku and Luna.

The startup, which emerged from stealth six months ago, was co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with participation from Nvidia and Jeff Dean. Initial targets include robotics firms, scientific labs, and media companies. Elorian aims to build the world’s best visual reasoning model next year, with long-term goals of enabling AI to control robots by building models of physical scenes.