Elorian introduced Glimpse on October 16, 2026, an AI model aimed at interpreting images and video. For analysts reading charts and media teams organising footage, it adds a candidate for tasks such as explaining a graph or labelling events in clips.
Elorian says Glimpse can revisit image regions and short video intervals while working out an answer. The potential benefit is closer inspection of an ambiguous detail, such as an object moving across a scene. That is a proposed advantage, not a guarantee of correct answers.
Together AI describes Glimpse as a research preview accepting text, images and video and returning text. Its listed output is an explanation or description, not a finished video.
What the launch evidence shows
In Elorian’s benchmark table, Glimpse scores above GPT-6 Astra on short-video captioning but below it on detailed video captioning. Results use different model configurations and remain vendor-reported, not independently verified comparisons.
The launch examples also distinguish recorded train and bird outputs from illustrative tire, railway-switch and pliers explanations. Those illustrations should not be read as successful model runs. Elorian identifies direct robot control as future work.
Token prices are only part of the bill
Elorian lists $0.60 per million uncached input tokens and $3 per million output tokens. Tokens are the billing units for what a model processes and produces.
Illustrative calculation: if a task used 10,000 billed uncached input tokens and 2,000 billed output tokens, those rates would give $0.006 + $0.006 = $0.012, or $12 for 1,000 identical tasks. These are assumed counts, not measured Glimpse usage.
This is not a price per image or minute of video. Additional charges, retries and staff checking are excluded; typical total costs and time savings remain unestablished.
Before trying it at work
Elorian advertises API access for connecting business software, plus chat, and support for company evaluations. Immediate access for any individual account is not established. Together lists dedicated deployment but also displays an upcoming-endpoint notice; confirm availability and pricing with the chosen provider.
Before uploading private charts or footage, check the applicable data settings and agreement. Elorian’s privacy policy says free-service inputs and outputs are used for training by default, with an opt-out; paid, API and enterprise terms may differ.
A useful evaluation would separate chart reading from video tagging. Check axis labels and units for charts, and missed events or wrong timestamps for video. Compare the full cost of a correct result, including human corrections, against the current workflow.
If the actual task is turning chart values into tables, RohitAI’s guide to chart and document extraction covers that different problem.