Dream-1 robot executing a task via Dream Labs' public API, illustrating zero‑shot control.
DREAM-1 LAUNCHES ZERO-SHOT ROBOT API WITH $90M FUNDING
*Dream Labs exits stealth with a $90 million seed round and a public API that lets any robot run tasks out of the box. The service charges $3 per hour and promises zero‑shot control of unseen hardware, aiming to break the long‑standing robotics chicken‑egg problem.*
Dream Labs has burst out of stealth with a $90 million seed round and a product that could rewrite robot economics. Dream-1, a bimanual robot with parallel grippers, is offered through a public API that charges $3 per hour and requires no fine‑tuning, demonstrations, or machine‑learning expertise. The claim is stark: control robots the system has never seen, in environments it has never visited, on tasks it has never been shown. If true, the service could dissolve the robotics chicken‑egg problem that has stalled mass adoption for decades.
The company’s narrative hinges on two converging trends. First, robot hardware has plummeted to $2 k for a capable two‑arm setup and under $15 k for semi‑humanoid platforms. Second, the World Action Model, unveiled in the February 2026 DreamZero paper, demonstrated zero‑shot generalization after just 30 minutes of robot data. Dream-1 is the first commercial incarnation of that model, promising immediate utility that fuels the data loop it once needed.
Can Dream-1 control unseen robots without training?
Dream-1 is a bimanual robot with parallel grippers that connects to Dream Labs’ public API. Users upload a robot description, pay $3 per hour, and the model issues commands without any fine‑tuning, demonstrations, or machine‑learning expertise. The system claims to operate on robots it has never seen, in environments it has never visited, and on tasks it has never been shown. This zero‑shot capability stems from the World Action Model introduced in the February 2026 DreamZero paper, which generalized after only 30 minutes of YAM robot data. The API is live worldwide, turning a $2 k hardware setup into an immediately useful tool.
Why has the robotics chicken‑egg problem persisted?
Robotics has long required data from a specific robot to make that robot useful, yet usefulness drives purchase, which generates data. The loop stalled adoption. Self‑driving cars broke a similar cold‑start by being useful with zero autonomy, gathering massive data from everyday drivers. Most robots lacked that baseline utility. Dream-1 attempts to invert the equation: a generalizable model provides immediate capability, prompting deployment, which then feeds new data back into the model. The approach mirrors the self‑driving breakthrough, but applies it to manipulation tasks across diverse hardware.
How cheap hardware is reshaping robot adoption
Hardware costs have collapsed faster than analysts expected. A capable two‑arm setup now costs as little as $2 k, and seven semi‑humanoid robots are available for under $15 k—a price tier that barely existed a year ago. Hugging Face’s $399 Microduck sold more than 10,000 units within four days of its late‑August launch, showing consumer appetite for affordable, delightful machines. These price points push robots into homes, labs, and small teams, creating a base of devices that can generate the data needed for models like Dream-1 to improve.
What real‑world tasks has Dream-1 already performed?
Beta testing spanned 12 sites across four continents. In Shenzhen, China, Dream-1 sorted candy; in Singapore it pulled a tissue; in Kuala Lumpur it stacked cups; in Paris it moved a red cube to a green container. Additional runs in Seoul folded a box, in Jakarta found a marker, and in San Francisco stood three bottles upright. Autonomous demonstrations included folding a dress, clearing dishes, pulling a highlighter cap, and scooping cat litter. Each task was executed on robots the model had never encountered, confirming zero‑shot embodiment generalization.
Frequently asked questions
What is the cost to use Dream-1's robot API?
Dream-1 is offered at $3 per hour, with no fine‑tuning or machine‑learning expertise required.
How much does a capable bimanual robot setup cost today?
A capable two‑arm setup can be purchased for as little as $2,000.
What evidence does Dream Labs provide for zero‑shot generalization?
DreamZero, released in February 2026, generalized to unseen tasks after post‑training on just 30 minutes of YAM robot data, and Dream-1 replicates this capability across multiple beta sites.
With $90 million in seed capital and a pricing model that undercuts traditional robot deployment, Dream Labs is positioned to accelerate a feedback loop that has eluded the field for years. If the zero‑shot promise holds across the growing fleet of $2 k arms and $399 consumer bots, the next wave of robots could become data generators as quickly as cars became drivers. The industry now watches whether Dream-1 can turn a theoretical breakthrough into a market‑wide catalyst.
Sources: Hacker News – Dream Labs research page (https://dreamlabs.ai/research/dream-1/)