DeepSeek founder Liang Wenfeng has recently been paying close attention to embodied AI and has held conversations with a number of founders in the field, Phoenix Tech reported on October 12. The discussions covered robotics, large models and how AI systems interact with the physical world.
According to people familiar with the matter cited in the report, Liang is considering whether DeepSeek should move further into embodied AI. Two routes are said to be under discussion: assembling an in-house team to do the research and development directly, or entering the field through stakes in companies already working on it. No decision has been reported.
One person told Phoenix Tech that Liang does not have much experience in the field and has been studying it intensively. DeepSeek had not responded to requests for comment as of publication, Sina Tech noted.
The report linked part of the interest to recent progress elsewhere. It said several test demos have shown OpenAI's GPT-6 achieving execution results on embodied tasks that surprised the industry, and suggested that this is one of the pressures behind DeepSeek's thinking.
Another industry figure quoted in the coverage said a move would not be surprising. Zhipu and Moonshot AI, the company behind Kimi, both already have embodied AI work in reserve, the person said, and embodied intelligence has become a new area of focus for AI companies worldwide.
For DeepSeek, robotics would be a significant expansion. The company has so far concentrated on language models and the software around them. In recent weeks it shipped DeepSeek-V4.1-Flash with native multimodal input, open-sourced its Harness agent runtime, and released Ascend versions of its core inference libraries. Liang has also said publicly that the company's next models must be trained on domestic chips.
Embodied AI asks for different capabilities: data gathered from physical robots, simulation, hardware partners and control models that can act in real time. Whether DeepSeek builds those itself or works through partners, the reported talks suggest the company is at least studying how its models could move off the screen and into machines. For now, the plans remain under consideration.
SEE ALSO: DeepSeek's Liang: Next Models Must Train on Huawei and Domestic Chips · DeepSeek Open-Sources Harness Agent Runtime