Hangzhou-based DeepSeek released the general-availability version of DeepSeek V4 Pro on Aug 13, while Alibaba, another Zhejiang-based technology company, launched its flagship Qwen3.8-Max model 10 days earlier.

According to DeepSeek's own benchmark results, V4 Pro showed significant gains, with some scores approaching those of leading proprietary models such as Anthropic's Fable 5. Qwen 3.8-Max ranked fifth in the Text Arena at launch, according to Alibaba.

DeepSeek V4 Pro has 1.6 trillion total parameters, with about 49 billion active parameters per inference, while V4 Flash has 284 billion total parameters and 13 billion active parameters. Both are open-weight and support context windows of up to 1 million tokens.

Qwen3.8-Max has 2.4 trillion total parameters and activates 95 billion parameters per inference. It uses a sparse mixture-of-experts architecture, is open-weight, and supports a context window of up to 1 million tokens, allowing developers greater flexibility to build, adapt, and commercialize applications based on the model.

A total of 90 generative AI services have completed regulatory filing in Zhejiang. Some were developed by adapting and fine-tuning foundation models such as Qwen and DeepSeek.

ModelScope, which operates a developer center in Hangzhou, has attracted more than 29 million AI developers worldwide and hosts over 200,000 models, helping lower barriers to AI application development.

Zhejiang is also making more computing power, data, and other resources available to developers.

Hangzhou aims to attract more than 1,000 companies in the open-model ecosystem and develop into a national hub for low-cost computing and model innovation.

Across the province, Zhejiang plans to accelerate the development of core AI industries including agents, application services, and intelligent devices.