AI startup Elorian has released Glimpse, a visual-reasoning model that can return to source imagery while solving a problem. During reasoning it can crop a region, retrieve selected video frames or run code to measure and compare what it sees, rather than converting an image into tokens once and relying only on that initial representation.
Glimpse is available as a research-preview API, while its chat interface remains waitlist-only. Elorian is initially targeting robotics companies, scientific laboratories, media groups and creative businesses with large image or video collections. The API costs $0.60 per million input tokens and $3 per million output tokens.
The company says it trained the model with a small team on a few hundred GPUs. Its own evaluations report stronger results than GPT-6 Astra on scientific-chart reasoning and short-video captioning, plus competitive performance on other visual tasks. Those are company claims, and independent testing is still needed. The basic technique is also not entirely new: OpenAI added crop-and-zoom operations to its reasoning models in 2025.
Elorian’s distinction is to center the model on repeated visual inspection. The preview will show whether that design helps on real tasks where diagrams, symbols or changing video details cannot be reliably summarized in one pass.