In brief
- Glimpse can revisit specific parts of an image while reasoning.
- The model can extract specific frames from video and run code to measure or compare images.
- Elorian announced a price of $0.60 per 1 million input tokens and $3 per 1 million output tokens.
- The chat interface is waitlist-only; the model is available via API as a research preview.
Palo Alto-based AI startup Elorian announced its first model, Glimpse, as a research preview on October 16, 2026. The model is available via API; access to its chat interface is currently limited to a waitlist. Elorian’s approach is built around revisiting the visual source while reasoning, rather than inspecting an image once and converting it into a textual representation.
How does Glimpse inspect images?
The method Elorian calls “visual scratch space” allows the model to crop and closely examine part of an image during its analysis. The system can select specific frames from video and run code to measure or compare what it sees. This lets the model recheck visual details instead of relying solely on the conclusion it reached at first glance.
One example Elorian CEO Andrew Dai shared with TechCrunch is that interpreting visual symbols without text can still be difficult for models. The company says Glimpse aims to reduce this limitation in visual reasoning tasks. However, the model’s use of this method does not mean it will perform equally well across every type of image or at every scale.
Preview, test results and pricing
Glimpse is available via API as a research preview. The chat interface is not yet generally available and can be accessed through a waitlist. Elorian announced pricing of $0.60 per 1 million input tokens and $3 per 1 million output tokens.
According to the company’s own evaluations, Glimpse outperforms GPT-6 Astra on reasoning about scientific charts and short-video captioning tests, and delivers competitive results on some other visual reasoning tasks. These performance claims are based on Elorian’s tests. The source does not include independent evaluation results or benchmarks specific to architectural visualization.
According to TechCrunch, Elorian says it trained Glimpse using a few hundred GPUs and that the model is significantly smaller than the frontier systems it was compared with. This information concerns the training process; it does not explain the hardware end users would need to run the model or the computational requirements on the API side.
What it means for architecture and visualization workflows
A model that can inspect image and video content may be worth testing in workflows where architecture and interior design teams work with visual material. For example, teams could evaluate how well the model identifies details in an image or assesses specific frames in a short video in their own use cases. These are potential areas for testing; Elorian’s announcement does not promise compatibility with architectural drawings, BIM files or specific design software.
For that reason, evaluation should consider not only general test results but also the types of visuals the model will be used with and the cost of the API. Since Glimpse’s chat interface is waitlist-only, teams looking to test it via API will need to review access and pricing conditions separately. The source provides no information about licensing terms, file format compatibility or the hardware users may need.
Sources
1 sourceSource texts are not republished; short quotes are marked, everything else is our own summary and commentary.
Glimpse’s ability to revisit visual sources and check details offers a different approach from systems that only describe an image. For architecture and visualization teams, the key measures will be how consistently this approach works on real project images and whether the API pricing suits their usage volume. There is not yet enough data to generalize the reported test results directly to architectural workflows.
Offices in Turkey should clarify access, licensing terms and compatibility with the file types they plan to use before testing the model. The number of GPUs used for training does not indicate local hardware requirements, and the source does not say whether the model can run on a computer. For now, a more cautious approach would be to limit initial evaluations to API access and sample tasks.
Frequently asked questions
How can Glimpse be used?
The model is available via API as a research preview. To access the chat interface, users need to join the waitlist.
How much does Glimpse’s API cost?
Elorian announced pricing of $0.60 per 1 million input tokens and $3 per 1 million output tokens.
Is Glimpse compatible with architectural drawings?
The source provides no information about compatibility with architectural drawings or specific file formats. Its performance on architectural images should be tested separately.
