Unitree Founder: GPT-6 Could Trigger a Robotics Breakthrough as Early as 2027

Wang Xingxing, founder and CEO of Hangzhou-based Unitree Robotics, said on Monday, October 12, that OpenAI's recently released GPT-6 model could become a turning point for robotics. In his words, expectations for a breakthrough have now shifted to a horizon of "one to two years." He allowed that a GPT-7-class model released in 2027 could directly affect the industry.

What exactly did the head of Unitree say?

The source, SCMP, quotes Wang Xingxing as saying: "Perhaps, say, a GPT-7 model released next year could directly..." The quote breaks off in the available excerpt, so we will not finish the thought for the speaker. We will record only what was stated definitively.

First, GPT-6 is already out, and Wang Xingxing cites it as the reason for his optimism. Second, he framed the timeline as a cautious range, not a date. Third, he was talking about the value of language and multimodal models for physical machines, not about a specific Unitree product.

Why this forecast deserves a close read

Forecasts of a "year of robotics" come around regularly, and most of them originate with investment funds and in slide decks. Here the speaker runs a company that sells robots in industrial volumes. Unitree is known for its inexpensive Go2 quadrupeds and its H1 and G1 humanoid platforms. At launch, the G1 was priced at roughly 16 thousand dollars, and the compact R1 was announced in 2025 for 39,900 yuan, about 5,900 dollars. Around the same time, the company was preparing to list on the Shanghai STAR Market, so it has a reason to show investors where the market is heading.

That is a reason to discount for the speaker's interest. But his argument is substantive: hardware is cheaper today than ever, while a robot's ability to act in an unfamiliar environment remains a weak spot.

Where robotics actually gets stuck

Mechanics, motors and batteries have become cheaper and more reliable in recent years. The difficulties lie elsewhere: generalizing to new tasks, manipulating small and deformable objects, and a shortage of high-quality data on physical interaction. A language model knows what "pour water into a glass" looks like, but it does not know how much force the fingers of a particular manipulator need. If GPT-6 and its successors plan well and understand a scene well, that is where they could give robots the most.

This is our analysis, not a claim by Wang Xingxing: he named a timeline and a trigger, and the mechanism of the breakthrough is not disclosed in the available text.

Key figures and facts:

• Timeline given by Wang Xingxing: "within one to two years."
• Trigger model: GPT-6, recently released by OpenAI.
• Hypothetical next step: GPT-7 in 2027.
• Unitree headquarters: Hangzhou, China.
• Unitree price benchmarks: G1 about 16 thousand dollars, R1 about 5,900 dollars at the time of announcement.

What teams and investors should do

Developers at robotics startups should start checking now how their stack connects to an external model: the interface between an LLM-based planner and the low-level controller should survive a model swap without a rewrite. If GPT-7 does arrive in 2027, the winners will be those who can quickly run the new version through their own scenarios.

For investors, the practical benchmark is this: look not at the number of humanoid robots on display, but at which tasks they perform in an unfamiliar environment without manual tuning. If such demonstrations become routine after GPT-6, Wang Xingxing's forecast will be borne out. For now, it is the opinion of a manufacturer with an obvious interest, not a measured result.