Elorian's Glimpse model reasons with images, not just about them
elorian releases glimpse, an ai model that can reason with images by revisiting and zooming into specific parts, outperforming gpt-6 astra on some tasks. this could change how ai interacts with visual data.
elorian, a palo alto ai startup, has released its first model, glimpse, which can reason about images by revisiting and zooming into specific parts. this is different from most multimodal models that look at an image once and then reason in text. glimpse uses a mechanism called visual scratch space, which allows it to crop into a region, pull specific frames from a video, or run code to measure or compare what it's looking at. elorian claims that glimpse outperforms gpt-6 astra on scientific chart reasoning and short-form video captioning. the model is launching as a research preview with api access and a chat interface is waitlist-only. elorian aims to use this technology to help companies with lots of video data, such as robotics firms, scientific labs, media and creative companies.