• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI
    Announcement

    Reka AI Labs releases Rho-1 research preview for multimodal AI and robot actions

    Reka describes one model for creating and reasoning across media, with unresolved limits in video consistency and editing.

    T(
    MA
    DY
    15 Sources, ,

    TLDR

    Reka AI Labs released Rho-1, a 19-billion-parameter research preview it says handles text, images, video and robot actions in one network. The company describes steerable video and shared state across conversations, and says it trained the model on 320 H100 GPUs in about three months. Its announcement also lists structural drift, unreliable object grounding across video, brittle edits and a 672-by-384 video resolution cap.

    Combined views

    27.7K

    15 Sources, first seen 2h ago

    Combined views

    27.7K

    15 Sources, first seen 2h ago

    442 likes

    Useful links

    reka.ai

    Reka Labs | Where Multimodal Reasoning Is Built
    2h ago
    first seen 2h ago
    442 likes
    35 comments
    152 saves
    68 reposts
    35 comments
    152 saves
    68 reposts

    16 Sources

    reka.aiRho-1: Collapsing the multimodal stack
    @RekaAILabsToday, we are releasing a research preview of Rho-1, our 19B omni model that understands and generates text, images, video and robot actions, all inside a single neural network.2h
    @DaniYogatamareally proud to share what the team has been working on. rho-1 is our 19B omni model. reasoning, real-time video simulation, and continuous robotic control inside a model. we are very early on the compute curve and candid about current limits in the post, but the compounding gains across modalities are already clear.2h
    @artetxemIncredibly proud of what the team built! Here are some unedited screen recordings of Rho-1, exactly as it ran. No agent calling specialist models behind the curtain. Just one model, end to end.2h
    @teortaxesTexvery ambitious omnimodel moonshot49m
    @altryne@RekaAILabs Whoah! 100% going to cover this on @thursdai_pod ! Congrats on the release!34m

    Reka AI Labs released a research preview of Rho-1 on Oct. 5. The company describes the 19-billion-parameter model as one neural network for understanding and generating text, images, video and robot actions.

    Featured Source

    Reka says it can create, edit and reason across media in one conversation, with each step drawing on shared state. The company also describes continuous video generation that accepts new instructions while running, changing the scene’s trajectory without a cut.

    For robotics, Reka says the same model weights predict future camera observations and direct joint movements, without a separate planner. Its announcement includes a LIBERO simulation example; that example does not establish performance on deployed robot hardware.

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Useful Links

    reka.ai

    Reka Labs | Where Multimodal Reasoning Is Built
    Today's Rank

    #3

    Today's Rank

    #3

    16 Sources

    reka.aiRho-1: Collapsing the multimodal stack
    @RekaAILabsToday, we are releasing a research preview of Rho-1, our 19B omni model that understands and generates text, images, video and robot actions, all inside a single neural network.2h
    @DaniYogatamareally proud to share what the team has been working on. rho-1 is our 19B omni model. reasoning, real-time video simulation, and continuous robotic control inside a model. we are very early on the compute curve and candid about current limits in the post, but the compounding gains across modalities are already clear.2h
    @artetxemIncredibly proud of what the team built! Here are some unedited screen recordings of Rho-1, exactly as it ran. No agent calling specialist models behind the curtain. Just one model, end to end.2h
    @teortaxesTexvery ambitious omnimodel moonshot49m
    @altryne@RekaAILabs Whoah! 100% going to cover this on @thursdai_pod ! Congrats on the release!34m

    Speed claims come with qualifications

    In its technical announcement, Reka reports median video generation at 0.79 times real time for the base model, with a viewable stream starting in about six seconds. It says a distilled variant produced a 5.3-second clip in about one second in internal testing. The company notes that generation waits in its conversational replay are shortened.

    Reka says the model was trained from scratch on 320 H100 GPUs in about three months.

    Where the preview falls short

    Reka’s limitations section says longer video sequences can preserve realistic textures while losing coherent room layouts. Tracking and locating objects across video remain unreliable, targeted edits are inconsistent across prompts, and native video resolution is capped at 672 by 384.

    The company presents Rho-1 as a proof of concept rather than a finished product.

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Useful Links

    reka.ai

    Reka Labs | Where Multimodal Reasoning Is Built