Models · 5 October 2026 · 1 min read
Reka Rho-1: one model for text, images and video
Previewed on 5 October 2026: one network that both understands and makes text, pictures and video, at low resolution for now.
By The Crewnema team
Reka announced Rho-1 on 5 October 2026. Most products that make video join a language model to a separate video model; Rho-1 does both in one network. These are the points from Reka’s announcement.
What Reka announced
- A 19B model trained from scratch that reads and generates text, images and video, and can also output actions for robots.
- A distilled version, Rho-1 Flash, that cuts generation from 99 steps to 8.
- Native video capped at 672 by 384 pixels.
- Known limits: long clips drift in structure, editing is inconsistent across prompts, and finding objects works on still pictures but not yet across video.
- A research preview, open to organisations by request.
Our view
Nothing here is usable for an advert today, and Reka does not pretend otherwise. What is interesting is the direction. A model that understands a picture and makes one in the same network could, in time, check its own work: notice that the label on the bottle changed and fix it. That is the job Crewnema does today with separate measuring steps.
Side by side
In Crewnema
Rho-1 is not available to use and is not on Crewnema’s roster. How Crewnema checks that a face and a product hold across shots is in keeping the same face and product in every shot.
Sources
Make your first video
Crewnema is in private beta and lets businesses in a few at a time. Request access and we will write when yours is open.