Reka AI releases research preview of Rho-1 omni-model
Reka AI unveiled Rho-1, a 19‑billion‑parameter omni‑model that can understand and generate text, images, video, and robot‑control actions within a single neural network. The model treats all modalities as tokens in one shared context window, eliminating the need for separate tool calls or external models. It can produce continuous video in real time and adapt to new instructions without…
Key points
- Reka AI unveiled Rho-1, a 19‑billion‑parameter omni‑model that processes text, images, video, and robot control in one network.
- The model was trained on 320 H100 GPUs for roughly three months, using an inverse‑dynamics approach to derive robot actions from internet videos.
- Rho‑1 follows Reka Core, the company’s April 2024 multimodal model that rivaled GPT‑4, Claude 3 and Gemini Ultra on benchmarks.
To compensate for limited robot‑training data, Reka AI built an inverse‑dynamics model that extracts control signals from ordinary internet videos. Rho-1 was trained on 320 H100 GPUs for about three months. The release follows the company’s April 2024 launch of Reka Core, a multimodal language model that performed competitively against GPT‑4, Claude 3 and Gemini Ultra on benchmark tests, reflecting a broader research trend toward unified “world models.”
Reka AI's omni-model Rho-1 handles text, images, video, and robot control in a single model
The Decoder · 5 October 2026
Loading the full article…
This text was published by The Decoder and written by Matthias Bastian. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Finetuned 1.5B Qwen to generate bash commands at gpt-4o level using 400k synthetic examples + Fully opensource finetune dataset · 1 src
- Xiaomi’s MiMo‑V2.6‑Pro tops open‑weight model rankings · 1 src
- Aleph Alpha releases Kolibri-1: 78B MoE model with 3.46B active parameters for German and English · 5 src
- TypeSafe AI's Jev decision model matches Claude Sonnet 5 accuracy at lower cost and latency · 4 src
- Yandex introduces Sona, a generative recommender that replaces recommendation cascade · 1 src
Comments
via GitHub Discussions