In May 2024, Unitree Robotics announced a humanoid robot priced at $16,000 and available for purchase. Not available "soon." Not available "to select partners." Available — as in, you could place an order. The G1, as the robot is called, ships to researchers, developers, and institutions that want one.
This is worth pausing on. The price of a Unitree G1 is roughly what you'd pay for a well-equipped used car. It is, by a significant margin, the lowest publicly stated price point for a general-purpose humanoid robot from any company currently shipping hardware. Boston Dynamics does not publish pricing for Atlas. Figure AI is not selling to the public. Tesla's Optimus has no announced price. Unitree has a price, and that price is $16,000.
What does that actually mean? What can the robot do for that money, and what does the price itself signal about the direction the industry is heading?
What the G1 Is
The Unitree G1 is a bipedal humanoid robot standing approximately 127 centimetres tall and weighing around 35 kilograms. It has 23 degrees of freedom — a measure of how many independently controllable joints the robot has — spread across its legs, arms, and hands. Its hands are three-fingered grippers capable of grasping objects, though the precision of that grasping varies considerably depending on the task and the object.
The G1 moves. It walks on uneven terrain, recovers from pushes, can squat, crouch, and navigate stairs at a cautious pace. Unitree's demonstration videos show it performing tasks including folding a cloth, opening a bottle, and picking up objects from the floor. These are real demonstrations of genuine capability — the robot is not being remotely operated frame by frame to simulate competence it does not have.
It is powered by a battery that provides roughly two hours of operation under normal conditions. It has an onboard computing unit that handles perception and locomotion, with the option to offload more demanding AI inference to external systems over a network connection. The robot is designed to be programmable — Unitree provides a software development kit and the G1 runs on a ROS2 (Robot Operating System 2) compatible architecture, which is standard enough that researchers familiar with other robotic systems can work with it without starting from scratch.
What It Can Actually Do — Honestly
The honest answer is: perform a narrow range of physical tasks reliably in structured environments, and a wider range of tasks unreliably in less structured ones.
Where the G1 is most capable is in locomotion — walking, navigating, balancing, recovering from perturbations. Unitree has been building quadruped robots (four-legged, dog-shaped machines) for several years, and the balance and movement engineering from that work carries over. The G1 moves with more fluency than you might expect from a $16,000 robot.
Where it is more limited is in dexterous manipulation — the precise use of hands to interact with objects. The three-fingered gripper design handles a reasonable range of everyday objects but struggles with anything requiring fine motor control, irregular shapes, or delicate handling. This is not a criticism unique to Unitree. Dexterous manipulation is one of the hardest unsolved problems in robotics, and no humanoid robot on the market today handles it reliably across the full range of tasks a human hand can perform.
The G1's AI capabilities are similarly bounded. It can be trained to perform specific tasks through reinforcement learning — a technique where the robot learns through trial and error, guided by a reward signal — but the resulting skills are narrow. A G1 trained to fold a specific cloth in a specific way does not automatically know how to fold a different cloth in a different orientation. Generalisation across tasks remains a genuine limitation, and any coverage that implies otherwise is not being straight with you.
The Price Signal
The $16,000 price point is the most analytically interesting thing about the G1, more interesting than any specific capability.
Humanoid robots have historically existed in one of two categories: research platforms costing hundreds of thousands of dollars and available only to well-funded institutions, or announced products with no actual price and no actual shipping date. The G1 breaks that pattern. It is a shipping product with a public price that falls within reach of a university lab, a mid-sized research team, or a well-funded startup.
How Unitree achieved this price is partly a function of being a Chinese manufacturer with access to lower-cost components, labour, and manufacturing infrastructure. The actuators — the motorised joints that allow a robot to move — account for a substantial portion of the cost of any humanoid robot, and Unitree manufactures its own, which reduces their cost compared to sourcing them externally. This is a supply chain advantage, not a technical shortcut.
The implication of a $16,000 humanoid robot is not that humanoid robots are now cheap. It is that the cost curve is moving in the direction that matters. If a robot that costs $16,000 today costs $8,000 in three years and $4,000 in five — which is not a guaranteed trajectory but is a plausible one given the dynamics of hardware manufacturing at scale — then the economics of humanoid deployment change substantially.
The comparison point that matters is not what a humanoid robot costs today versus what a human worker costs today. It is what a humanoid robot capable of performing a given task will cost in five years, versus what that task costs in human labour at that time. The G1 is an early data point in that calculation, not the conclusion.
Who Is Actually Buying One
The realistic buyer of a G1 today is a researcher, a robotics team at a technology company, or an institution with a specific application it wants to test. It is not a warehouse operator, a manufacturer, or anyone who needs a robot to perform production work reliably and unsupervised.
That matters because the G1's current utility is primarily as a research and development platform — a way for teams to work with humanoid hardware at a price point that doesn't require securing a seven-figure capital budget first. More teams working with humanoid hardware means more data, more trained models, more learned manipulation skills, and faster iteration on the software side of the problem. The hardware's limitations don't disappear, but the pace at which those limitations are addressed accelerates when the hardware is accessible.
There is also a second category of G1 purchasers worth noting: companies and research groups outside China and the United States — in Europe, South Korea, Japan, and elsewhere — for whom the G1 represents the most practical route to having humanoid hardware in-house. The competitive geography of humanoid robotics is not just about which robot is most capable; it is also about which robots are accessible, and at what cost, to teams in different parts of the world.
What to Watch For
The G1's significance will become clearer over the next eighteen months as researchers publish results from working with it. The relevant questions are not whether the robot can fold a cloth in a controlled demonstration — it can — but whether teams working with it at scale discover capabilities or limitations that change the picture of what this class of hardware can do.
Unitree has also announced the H1, a taller and more capable humanoid robot priced higher than the G1, which suggests the company is building out a product line rather than treating the G1 as a one-off. Whether that trajectory produces robots with meaningfully broader deployment utility depends on engineering progress that has not yet happened, in a field where progress is real but timelines are consistently uncertain.
The $16,000 humanoid robot is not the arrival of the humanoid workforce. It is, more precisely, the point at which humanoid robotics became accessible enough for a much larger group of people to start figuring out whether it can become that.