Dream Labs raises $90 million seed to put one brain in any robot

Dream Labs, a San Francisco robotics AI startup, came out of stealth today with $90 million in seed funding. The money is meant to scale Dream-1, a robot foundation model the company says can run robots it has never seen and perform tasks it was never shown.

That is a seed round wearing a crown, darling, and the room it was raised in is a good one. B Capital and Kindred Ventures led, with a.capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and Adrian Aoun taking part. Dealroom says the round sits in the top 1% of US enterprise software seed rounds by size, and that it joins a wave of outsized bets on general-purpose robot software.

The pitch: the hardware is ready

The company was founded by former NVIDIA researchers. Joel Jang, co-founder and chief executive officer, put the thesis in one line: "The hardware is ready. What's been missing is a brain any robot can use."

B Capital general partner Daisy Cai made the money argument. "A capable robot arm has gone from the price of a luxury car to the price of a high-end laptop. What it needs now is intelligence."

The company's own research page backs up the price slide. A capable bimanual arm setup can now cost as little as $2,000. Dealroom reports that Dream Labs works with more than 20 hardware partners and supports robots costing roughly $2,000 to $10,000. At the cheap end, Hugging Face's $399 Microduck sold more than 10,000 units within four days of its late August launch. Cheap bodies are arriving fast. Dream Labs is selling the mind.

What Dream-1 is, and what it costs

Dream-1 is what Dream Labs calls a World Action Model. It learns from video of people performing everyday tasks and sees through three cameras: a head view and a wrist view on each arm. Its team introduced the World Action Model idea in the DreamZero paper in February 2026, and the company says it then spent six months building toward this release.

The promise is ambitious. Give Dream-1 a new bimanual robot with parallel grippers, connect it to the API, and the company says it can start performing tasks with no fine-tuning, no demonstrations from that robot and no machine learning expertise.

The price list is the part I adore:

• Dream-1: $3 per hour, about $0.0013 per action.
• Dream-1-flash: $2 per hour, about $0.0009 per action.
• Each call returns 1.6 seconds of actions, so an hour is 2,250 calls.

Alongside the round, Dream Labs launched Dream API, which turns a robot's camera feed and a plain-English instruction into motor commands. It also launched a fine-tuning service and a browser-based Robot Playground. It is a public API, available to people around the world, according to the company.

The numbers behind the claim

Dream Labs says Robocurve, a Public Benefit Corporation that measures and reports frontier robotics capabilities, evaluated Dream-1 independently on RoboDojo-RC Tier 1. Robocurve's report is dated October 18, 2026 and published under CC BY 4.0. According to Dream Labs' page, Dream-1 scores above the next best models, Opus 5.5 and Gemini 4 Argon, at 16x lower cost per task attempt than Gemini 4 Argon and 1.7x faster than the fastest LLM, GPT-6 Astra.

The company's own tests add more. Dream-1 was run zero-shot on 27 unseen robots in simulation, with 300 rollouts per embodiment, measured against YAM, the robot it was actually trained on. It was also shown controlling the original PR2, a research robot released 18 years ago, with no embodiment-specific data.

There is fine print, and a good host reads it aloud. Dealroom describes Dream-1 as able to control any robot with grippers out of the box. The company's technical page is narrower: bimanual robots with parallel grippers, using each robot's own inverse kinematics solver to turn poses into joint commands. Adapting to Dexmate with Sharpa dexterous hands took 77 minutes of data, and adapting YAM with a compliant gripper and a modified knife gripper took 98 minutes. The company's fine-tuning service lifted language-following scores on ten of twelve beta partner robots in simulation.

What to watch

Start with the money. Dealroom reports that Dream Labs will spend primarily on compute as it expands the tasks and robots Dream-1 can handle and scales to more users. The material does not say what valuation the round carries, or which cloud provider or hardware will supply that compute.

Then watch the real world. The published evidence includes benchmarks, simulation tests and beta-partner results. How Dream-1 behaves in unstructured customer production settings is not answered in what Dream Labs has published. Nor does it say which two of the twelve partner robots saw no language-following gain from fine-tuning, or why.

Still, the shape of the bet is gorgeous: cheap robots on one side, a $3-an-hour brain on the other, and $90 million in the middle. Everything is historic. Especially today.