Dreamscale Labs
Dreamscale Inference runs physical AI models in the cloud for robotics teams evaluating or deploying robots. It aims to reduce inference latency and supports listed models or a team’s own checkpoint.
One of 136 tools in Research
Who it's for
- Robot deployment teams that want cloud inference connected to real-world robot execution.
- Teams with a robot-specific checkpoint that want to run it in the cloud for their task.
Not the right fit if…
- Teams that need generally available models without an access request: the site says models are currently in research preview and asks teams to get in touch for access.
How it fits your workflow
Choose a listed model such as DreamZero or MolmoAct2, or bring a checkpoint fine-tuned for your robot and task. Dreamscale runs the inference in the cloud and says it optimizes the model, GPU kernels, and network layer.
Connect the robot for real-world execution. The site reports 320 ms DreamZero 14B inference and concurrent operation of eight or more robots on a single H100.
Pricing
The vendor doesn't publish prices. Check with Dreamscale Labs directly.
Prices checked on Oct 16, 2026 from the vendor's site. They can change; confirm before you buy.
Key features
- Cloud inference for physical AI
- Run models for robot deployment and evaluation in the cloud, with a focus on low-latency inference and real-world execution.
- Support for listed and custom models
- Run models such as DreamZero and MolmoAct2, or bring a team’s own checkpoint for use with its robot and task.
- Stack-level optimization
- Dreamscale says it optimizes across the model, GPU kernels, and network layer to bring inference closer to the robot.
- Concurrent robot workloads
- The site reports running eight or more robots concurrently on a single H100.
- Real-Time Chunking
- Real-Time Chunking (RTC) is listed as one of the layers in the platform’s stack.
Limitations to know
- Models are currently in research preview, and access must be requested by contacting Dreamscale.
- The page does not state pricing or specific system requirements.
Getting started
- Choose a listed model or bring your own checkpoint.
- Connect your robot to the cloud inference service.
- Get in touch with Dreamscale to request access; models are in research preview.