Skip to content
Neural Daily

ModelsNews

Dream-1 ships a robot API with a $90M seed

On October 19 Dream Labs exited stealth with a $90 million seed led by B Capital and Kindred Ventures. It also opened a public API for Dream-1, a World Action Model that can control unseen bimanual robots with parallel grippers without fine-tuning.

By: Mikola Máté Patrik · · 2 min read

Still from the Dream Labs launch video: a presenter in a lab with robot arms in the background
Image: Dream Labs · Press kit · Official share image of the Dream-1 announcement · Source

What Dream-1 does

Dream-1 is a World Action Model. It learns video and actions together, so controlling a robot and predicting what it will look like next are the same problem. It sees through three cameras: a head view and a wrist view on each arm. It can also predict end-effector poses (gripper position and open/closed), which each robot's own IK solver turns into joint commands. That shared action space covers many different bodies (Dream Labs).

The company says the model zero-shots many unseen bimanual parallel-gripper robots, including the 18-year-old PR2. Pricing is $3 per hour for Dream-1 and $2 for Dream-1-flash. Each call returns 1.6 seconds of actions.

Independent eval and limits

Robocurve, a public-benefit corporation, evaluated Dream-1 on RoboDojo-RC Tier 1. Dream Labs says the model is better, faster and cheaper than the next best models (Opus 5.5, Gemini 4 Argon), 16× cheaper per task attempt than Gemini 4 Argon and 1.7× faster than the fastest LLM there (GPT-6 Astra).

Predicting end-effectors trades some control for generality versus joint angles. IK quality matters. The model does not control how the arm reaches a pose near obstacles or joint limits. Dream-1 is a research beta and is not validated for safety-critical use.

Fine-tuning as a Service

When zero-shot is not enough, FaaS adapts the model from a few minutes of partner demonstration data. On two partner robots, five minutes of data already improved results. Farther embodiments, such as dexterous hands, needed 77-98 minutes. Seed capital goes mainly to compute. More than 20 hardware partners are onboard, with supported robots priced roughly $2,000-$10,000 (Business Wire).

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles