China turns up the heat with open model blitz as US model makers panic
For the first time, Alibaba's Qwen team is letting its 'Max' model out of the API pen; meanwhile, DeepSeek V4-Flash gives new meaning to cheap and cheerful competition
On Monday, the AI arms race escalated as Alibaba, a Chinese e-commerce and cloud provider, unveiled Qwen 3.8-Max, a powerful 2.4 trillion-parameter model that competes directly with industry giants like Anthropic and OpenAI. This new model was introduced just days after DeepSeek V4 Flash 0731 made its debut, boasting performance on par with OpenAI's budget-friendly GPT-5.6 Luna at a 40% lower cost per task. With its 284 billion parameters, Qwen 3.8-Max is small enough to run on typical enterprise servers and workstations.
Chinese tech giants, including Alibaba, Moonshot, and DeepSeek, are now targeting their American counterparts by emphasizing both price and performance. This aggressive strategy comes as US model developers, such as Anthropic and OpenAI, intensify their concerns over the origin and safety of Chinese AI models. Anthropic CEO Dario Amodei emphasized his disapproval of open models originating from China and those that fall short of stringent safety metrics, effectively targeting anything that competes with his own models.
Clément Delangue, CEO of Hugging Face, the world's largest and most influential model repository, acknowledged China's dominance in the open weights arena, stating that they are likely to maintain their leading position in the next year. The United States currently possesses the most capable open weights model, Inkling, but it is significantly outperformed by models like DeepSeek's and Alibaba's Kimi K3 and Z.ai's GLM 5.2.
Alibaba, often compared to its American counterparts, has made its most powerful model, Qwen 3.8-Max, available for download for the first time. The model's multi-modal mixture of experts (MoE) architecture, with 95 billion parameters used for generating tokens, allows it to support context windows up to 1 million tokens. Qwen 3.8-Max is currently available via Alibaba's API service, QwenCloud, at $2 per million input tokens and $6 per million output tokens.
DeepSeek has also recently launched a more cost-effective alternative to OpenAI's GPT-5.6 Luna, further intensifying the competition in the open model arena.
Written by urgent.news from The Register's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.
