Elorian launches Glimpse, a visual thinking model, at $0.60 / $3 per million tokens

Elorian, the visual-AI startup that former Google DeepMind researchers launched in April, released its first model on October 16. Glimpse is what Elorian calls a visual thinking model: it reasons over images and video directly instead of first turning them into text. It is a research preview, available through Elorian's API and a chat app today.

THE MODEL

Elorian says Glimpse follows objects through motion and changing viewpoints, works out how parts fit together, and predicts what happens next in a scene. As an agent it runs code, transforms images and watches video with its own tools, and it can split a problem across subagents. Elorian says it built Glimpse with a small team and a few hundred GPUs by post-training an open-weights base model; it does not say which one, and Glimpse's own weights are not public.

Elorian's API docs list two model ids: elorian-1 and elorian-1-agent, the version that runs Python and searches the web on its own.

THE PRICE

$0.60 per million input tokens, $0.06 for cached input and $3 per million output tokens, on Elorian's platform. There is also a free chat tier with a weekly usage limit. Glimpse is not on OpenRouter yet.

HOW IT SCORES

Only Elorian's own figures so far. Its table puts Glimpse against GPT-6 Astra on 11 shared evaluations: Glimpse is ahead on four (WGO Bench robot manipulation, CharXiv-R chart reasoning, UGC-VideoCap captioning and FSC-147 counting), tied on ERQA and behind on six, including Video-MME v2 (56.7 against 64.1) and HSS (46.4 against 53.6). HSS is Elorian's own benchmark, built with Scale AI and published on October 8. No third-party board has scored Glimpse yet.

WHAT WE DO NOT KNOW

Which open model Glimpse is built on, its context window, and how it does on independent boards. "Research preview" also means the model or its price can change.