Dream Labs Raises $90M Seed, Opens Dream-1 Robot API at $3 an Hour

Dream Labs came out of stealth with a $90 million seed round led by B Capital and Kindred Ventures, and opened a public API for Dream-1, a robot foundation model priced at $3 per robot-hour ($2 for Dream-1-flash), according to its Business Wire announcement and launch post. The company says Dream-1 can run two-armed robots with parallel grippers that it was never trained on, with no data from that robot.

The numbers on the wire

• $90 million seed, led by B Capital with Kindred Ventures; A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and Adrian Aoun also participated. No valuation was disclosed.
• $3 per hour for Dream-1 and $2 per hour for Dream-1-flash. Each call returns 1.6 seconds of actions, so an hour is 2,250 calls, about $0.0013 per action on Dream-1 (company figures).
• 27 unseen robot bodies tested in simulation, with 300 rollouts per robot, plus real-robot runs on arms including the 18-year-old Willow Garage PR2.
• 20+ hardware partners; supported robots cost roughly $2,000 to $10,000.
• Fine-tuning: 5 minutes of a robot’s own data improved two partner arms; a dexterous-hand setup took 77 minutes.
• The company says the money goes primarily to compute.

How it works

Dream-1 is what the team calls a World Action Model: it predicts video and robot actions together, building on the DreamZero paper the founders published in February 2026. Dream Labs was founded this year by former leads of NVIDIA’s GR00T robotics research. Instead of outputting joint angles for one machine, Dream-1 can output where each gripper should be and whether it is open or closed, and the robot’s own inverse-kinematics solver turns that into joint commands. Because every robot looks like pixels to the head camera, the company argues video is the one representation all robots share.

The launch post is candid about limits. Zero-shot results depend on each robot’s IK solver, robots far from the training data can get infeasible predictions, and one-shot fine-tuning still beats zero-shot. Dream-1 is labeled a research beta that “has not been validated for safety-critical use.”

The benchmark claim

Dream Labs cites an independent pre-launch evaluation by Robocurve, a public benefit corporation, on its RoboDojo-RC Tier 1 suite dated October 18, 2026. Per the company’s summary, Dream-1 scored above Opus 5.5 and Gemini 4 Argon at 16x lower cost per task attempt than Gemini 4 Argon and ran 1.7x faster than GPT-6 Astra. That comparison pits a robot-control model against general LLMs driving robots, so it says more about cost and latency than about head-to-head robotics leaderboards.

What we are watching

Paid usage on the API beyond beta partners, Robocurve’s full report and whether other robot models get scored on it, and the promised support for wheeled mobile robots and dexterous hands.