Chinese AI models are pushing harder into the frontier race after DeepSeek released V4 for general availability and Alibaba unveiled its open-weight Qwen 3.8 model.
Both companies claim their latest systems can approach leading models from OpenAI and Anthropic. At the same time, they are using aggressive pricing and domestic hardware efforts to strengthen their position in the global AI market.
The announcements arrived within hours of each other over the weekend, highlighting China’s growing ambition to compete with the most capable AI systems available.
DeepSeek V4 Moves Into General Availability
DeepSeek made V4 generally available on July 19, around three months after its preview launch.
Early testers have reported strong performance in coding and reasoning tasks. During the preview period, V4 achieved an 80.6% score on SWE-bench Verified, a benchmark that evaluates how effectively AI models solve real software-engineering problems.
The company claims V4 can perform close to leading frontier systems while costing significantly less to use. However, its latest performance claims have not yet been confirmed by independent benchmark leaderboards.
DeepSeek also introduced dynamic pricing for V4. During off-peak periods, its V4 Pro model costs $0.87 per million tokens. That price is far below the reported cost of comparable usage on some premium US-based AI models.
Alibaba Unveils Qwen 3.8
Shortly after DeepSeek’s announcement, Alibaba introduced Qwen 3.8, a new open-weight model with 2.4 trillion parameters.
That scale would make Qwen 3.8 one of the largest open-weight AI models released so far. Open-weight models allow developers and organisations to inspect, adapt, and deploy the model more freely than closed systems.
Alibaba has made a preview of Qwen 3.8 available through its Token Plan and Qoder coding platforms. The company claims the model is close to the strongest frontier AI systems, although independent benchmark results are not yet available.
The preview is also priced at around one-tenth of Alibaba’s standard rates. Therefore, the strategy appears clear: offer high-end AI capabilities at a considerably lower price.
Lower Costs Put Pressure on US Providers
The latest Chinese AI models are not competing on benchmark scores alone. They are also challenging major providers on cost.
DeepSeek previously introduced a temporary 75% discount for V4 Pro. The company later made the lower price permanent, increasing pressure on services from OpenAI, Google, and Anthropic.
This approach could make advanced AI more accessible for developers and businesses with high usage volumes. It could also force frontier-model providers to reconsider their own pricing structures.
Meanwhile, Chinese startup Moonshot recently released Kimi K3, a 2.8-trillion-parameter open-weight model. The company also claims its system performs close to leading frontier models.
Together, these releases suggest that competition among Chinese AI labs is accelerating quickly.
Domestic Chips Are Becoming More Important
China’s AI push also includes efforts to reduce dependence on foreign chips.
DeepSeek reportedly optimised the V4 preview for Huawei’s Ascend hardware rather than Nvidia processors. The company is also believed to be developing its own inference chip, which could reduce its reliance on both Nvidia and Huawei.
Alibaba is following a similar path with its Zhenwu M890 AI chip.
US export controls have limited China’s access to some advanced AI hardware. As a result, Chinese companies are investing more heavily in domestic chips, software optimisation, and open-weight model development.
The latest releases show that Chinese AI models are becoming more capable, more affordable, and increasingly independent from Western technology.
0 responses to “Chinese AI Models Challenge Frontier Rivals With New Releases”