Chinese artificial intelligence companies DeepSeek and Moonshot AI reportedly relied on outputs from OpenAI’s GPT-6 Astra to help train their latest AI models, bringing renewed attention to a controversial technique known as model distillation.
The reported development highlights intensifying competition between US and Chinese AI developers, as companies seek ways to build capable systems while controlling the substantial costs associated with training advanced models.
According to a report published on October 28, 2026, by a technology news outlet and subsequently reported by Reuters, people familiar with the matter said both Chinese companies used outputs generated by GPT-6 Astra during the development of their latest models. The companies involved had not immediately responded to requests for comment when the report was published.
Which AI Models Are Involved?
The report identified two models that reportedly benefited from the technique.
Moonshot AI’s Kimi K3.1 Pro, released the previous week, was reportedly developed using outputs from GPT-6 Astra. DeepSeek’s upgraded V4.1-Flash 1017 model, launched on October 17, was also said to have benefited from the approach.
The available reporting does not specify the exact volume of AI-generated material used, the precise training process or how much the technique contributed to either model’s performance. The claims are based on people familiar with the development work rather than public technical disclosures from the companies.
Neither company had publicly confirmed the reported use of Astra’s outputs at the time of publication.
What Is Model Distillation?
Model distillation is a machine-learning technique in which one AI model learns from the outputs of another. Developers can use responses generated by a more capable model as training examples for a separate system, helping it reproduce selected capabilities without necessarily repeating the same expensive training process.
The technique has legitimate applications. AI developers routinely distil their own models to create smaller, faster and less expensive versions that can run more efficiently.
Controversy arises when a company uses a competitor’s proprietary model to reproduce its capabilities in ways that may violate the model provider’s terms or restrictions. Leading AI developers have argued that unauthorised distillation can allow rivals to benefit from substantial research and infrastructure investments without incurring comparable development costs.
Whether a particular case breaches contractual terms or applicable law depends on the circumstances and evidence. The reported use of a model’s outputs does not, by itself, establish that a legal violation occurred.
OpenAI Has Raised Concerns About Distillation
OpenAI has previously expressed concern about attempts to extract capabilities from its advanced models. In a September 30, 2026, blog post, the company said it had linked a significant part of a distillation campaign that began in July to individuals associated with Moonshot AI.
OpenAI has also said that it takes such activity seriously and continues to improve its safeguards as the methods used to access and reproduce model capabilities evolve.
Other AI developers, including Anthropic and Google, have similarly raised concerns about competitors using their systems to accelerate development. The debate has become more prominent as the quality of AI models improves and the financial stakes surrounding their development increase.
Why the Issue Matters for the AI Industry
Training frontier AI models requires substantial computing resources, specialised hardware, large datasets and experienced research teams. Distillation can potentially reduce the amount of work needed to reproduce selected capabilities, making it an attractive option for developers seeking to lower costs or accelerate releases.
At the same time, heavy reliance on another company's model outputs raises questions about intellectual property, commercial terms and fair competition. Companies developing proprietary systems have incentives to protect their investments, while the wider AI industry continues to debate how model-generated information can be used in training.
The issue also intersects with US-China technology competition, in which access to advanced computing hardware and powerful AI systems remains strategically important.
Questions Remain Unanswered
The latest report adds to scrutiny of how AI developers obtain training data and build competitive models. However, important details remain unverified, including the quantity of GPT-6 Astra outputs used, the precise methods involved and the effect on the final models.
The responses of OpenAI, DeepSeek and Moonshot AI could provide further clarity. Until additional evidence or official statements emerge, the reported use of distillation should be treated as an allegation based on sources familiar with the development work, rather than a fully documented account of the training process.
The episode underscores a central challenge for the AI sector: encouraging innovation and competition while establishing clear boundaries around the use of proprietary models and their capabilities.
