Elorian launches Glimpse, a model that re-examines images as it thinks
A new visual reasoning model beats GPT-6 Astra on four benchmarks and matches it on a fifth, priced 94% below Astra at $0.60 per million input tokens.
Elorian launched Glimpse on 16 October, its first model and a new category it calls a visual thinking model: one that reasons about images and video by returning to the source material mid-thought rather than converting everything to text up front. The model keeps what Elorian calls a visual scratch space, so it can zoom into a region, replay a short interval at a denser frame rate, or track one feature across time while it works.
On Elorian’s own benchmarks, Glimpse beats GPT-6 Astra on scientific chart reasoning (92.90 vs 92.10 on CharXiv-R), short-form video captioning (82.66 vs 80.32), open-vocabulary counting (83.36 vs 82.02), and manipulation-event understanding (42.70 vs 39.42), and matches it on embodied reasoning. Against Claude Opus 5.5, it comes out ahead on manipulation events and video captioning, though Opus leads on chart reasoning. Pricing is $0.60 per million input tokens and $3.00 per million output tokens, which Elorian says is 94% below Astra’s rate. It’s available now through the API and a chat UI on Elorian’s platform.
Why it matters: if you’re building anything that has to count objects, read a chart, or pull facts out of video, this is a frontier-level option at a fraction of GPT-6 Astra’s price, from a team with real backing (a $55M seed at a $300M valuation, Nvidia among the investors).
The caveat: every comparison here is Elorian’s own evaluation against other vendors’ models, not an independent benchmark. There’s no outside leaderboard entry yet to check the numbers against.