Key facts
- Glimpse can crop into regions, pull specific frames from video, or run code to measure or compare visual data.
- Elorian claims Glimpse is substantially smaller than frontier systems.
- Glimpse costs $0.60 per million tokens in and $3 per million tokens out.
- Elorian aims to build the world's best visual reasoning model next year.
Elorian, an AI startup that secured $55 million in seed funding, has launched its first model, Glimpse. The model, released on Friday, is designed to improve visual reasoning by allowing AI to revisit and interact with image and video data during its analysis, a capability Elorian calls a "visual scratch space." This approach differs from many multimodal models that convert visual data into tokens for text-based reasoning. CEO Andrew Dai stated that current AI models struggle with interpreting visual languages and symbols, citing an inability to determine the winner of a board game from a photograph. Elorian claims Glimpse was trained on a few hundred GPUs and is significantly smaller than competing frontier systems. The company also highlighted that Glimpse outperforms models like GPT-6 Astra on scientific chart reasoning and short-form video captioning, according to Elorian's evaluations. The initial target markets for Glimpse include robotics firms, scientific labs, and media companies, with future ambitions to enable AI to perform actions based on visual scene understanding.