China turns up the heat with open model blitz as US model makers panic
Chinese AI developers, including Alibaba, DeepSeek and Moonshot, launched a wave of powerful new open-weight models this week, intensifying pressure on American rivals like Anthropic and OpenAI. Alibaba released Qwen 3.8-Max, a 2.4 trillion-parameter model that rivals top offerings from its US competitors, marking the first time it has made its flagship "Max" model weights publicly downloadable rather than restricting it to a paid API. The release underscores growing concern that China now leads the open-weights AI race, even as US firms increasingly emphasise price, performance and safety concerns around Chinese-made models.
The surge follows DeepSeek's V4 Flash 0731, a 284-billion-parameter model that independent benchmarking firm Artificial Analysis found performs within a single point of OpenAI's budget GPT-5.6 Luna model while costing 40 percent less to run, and is compact enough for modest enterprise hardware. Hugging Face chief executive Clément Delangue told CNBC that Chinese firms are "clearly dominating" open models and could lead at the frontier level within a year. By contrast, America's leading open-weights model, Inkling, has under a billion parameters and lags behind Chinese offerings such as Moonshot's Kimi K3 and Z.ai's GLM 5.2, leaving Chinese models as the main open alternative to proprietary Western systems.
In short
- Alibaba releases Qwen 3.8-Max, its first open-weight flagship model
- DeepSeek's new model matches GPT-5.6 Luna at 40% lower cost
- Hugging Face CEO says China now dominates open-weight AI models
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Open-weight models represent a genuine advancement for the global AI ecosystem, allowing researchers, smaller enterprises and developers worldwide to access frontier capabilities without depending on expensive proprietary APIs controlled by a handful of Western firms. The performance metrics demonstrate competitive quality at substantially lower operational costs, creating real value for users whilst simultaneously spurring healthy innovation as all developers race to improve efficiency and capabilities. Open systems enable transparency, independent auditing and collaborative improvement—allowing the research community to identify and address safety issues collectively rather than trusting opaque proprietary systems from any single nation or company.
The case against
The rapid expansion of Chinese open-weight models raises justifiable concerns about national security, data sovereignty and geopolitical competition, particularly given the potential for influence from the Chinese government over systems made by domestically-based firms. Wide distribution of powerful open models without adequate safeguards or transparency about training data and alignment creates genuine risks of misuse at scale, and the attractive efficiency metrics should not eclipse the importance of careful evaluation of governance, oversight and long-term strategic implications. American technological leadership depends substantially on maintaining competitive advantage in AI development and the credibility to shape global governance standards; conceding the open-weights space to state-backed Chinese competitors risks surrendering market position and influence over how the world's most powerful AI systems are developed and deployed.
Explained simply
AI models are systems trained to generate text, answer questions and perform other tasks. Some are offered only through a company’s online service, while “open-weight” models let others download the trained model and run or adapt it themselves; that can make them easier to customise, though using them still requires computing power.
Chinese developers such as Alibaba, DeepSeek and Moonshot compete with US companies including OpenAI and Anthropic. Their models are attracting attention because they offer another route for businesses and researchers who want capable AI without relying solely on a company’s paid service.
The competition matters because access to strong models can affect the cost, spread and development of AI. It also raises questions for organisations choosing what to use, including how models perform, what they cost to run and how they handle safety and data.