Palo Alto-based AI startup Elorian has launched Glimpse, its first artificial intelligence model designed to analyze images and videos through repeated visual inspection rather than relying primarily on text-based reasoning. Released on October 16, 2026, the model is available as a research preview through an API and chat interface, following the company’s $55 million seed funding round at a $300 million valuation.
According to Elorian’s official announcement, Glimpse uses a visual reasoning process that allows it to examine specific image regions, revisit video frames, and run code while solving problems. The company says this approach improves performance on tasks involving scientific charts, object movement, spatial relationships, and video interpretation.
How Glimpse Uses Visual Reasoning to Analyze Images
Conventional multimodal AI models typically process images into tokens before reasoning about their contents using language. Glimpse takes a different approach by repeatedly examining visual information throughout its reasoning process.
Elorian calls this mechanism a visual scratch space, which allows the model to crop images, zoom into specific details, and review selected video segments. It can also execute code to measure objects, compare visual elements, and verify observations before generating responses.
CEO and co-founder Andrew Dai, a former Google DeepMind researcher, said the technology addresses limitations he encountered when testing AI models on everyday visual tasks. He cited an example in which a frontier model struggled to interpret the symbols and scoring system of a board game from a photograph.
The company developed Glimpse using a relatively small research team and several hundred GPUs. Rather than building a large general-purpose system, Elorian focused on improving visual interpretation while reducing training and operating costs.
Glimpse Reports Competitive Performance Against Larger AI Models
Elorian’s internal evaluations indicate that Glimpse outperformed GPT-6 Astra on selected benchmarks covering scientific chart reasoning, short-form video captioning, object counting, and robotics-related tasks. The company also reported competitive results against other frontier AI models, although the findings have not been independently verified.
Glimpse costs $0.60 per million input tokens and $3 per million output tokens. Elorian says these rates are 94% lower than GPT-6 Astra’s published pricing, making the model potentially suitable for businesses processing large volumes of visual information.
The company also acknowledged that repeated visual inspection can increase the number of tokens required for individual tasks. Improving token efficiency remains one of its development priorities.
Elorian Targets Robotics, Scientific Research and Media
Elorian emerged from stealth earlier in 2026 after securing $55 million in seed funding co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital. The investment round also included Nvidia and Google chief scientist Jeff Dean.
The startup is initially targeting robotics companies, scientific laboratories, media organizations, and creative businesses that work extensively with images and video. Its longer-term objective is to develop AI systems capable of interpreting physical environments and selecting appropriate actions, including controlling robotic equipment.
Dai said the company aims to improve Glimpse’s performance across visual reasoning benchmarks throughout 2027. Elorian plans to build on its current model to support increasingly complex tasks involving physical objects, movement, and real-world environments.
Featured image credits: Magnific.com
For more stories like it, click the +Follow button at the top of this page to follow us.