One Year After DeepSeek, Chinese AI Firms Accelerate Model Launches to Close Gap With U.S. Rivals

Just over a year after DeepSeek jolted global markets with a low-cost chatbot that challenged U.S. leaders, Chinese artificial intelligence companies are moving faster than ever to release new models and expand adoption at home and abroad. The renewed push highlights intensifying competition with American players such as OpenAI, Anthropic and Google, even as geopolitical and technology restrictions persist.

This week, Beijing-based startup Moonshot AI unveiled its Kimi K2.5 model, claiming advances in video generation and so-called agentic capabilities—systems designed to carry out tasks autonomously with minimal human input. The release came only months after the company launched its previous K2 model, underscoring how quickly development cycles are shortening among Chinese AI firms.

Around the same time, e-commerce heavyweight Alibaba introduced a new generative model under its Qwen lineup, capable of producing text, images and video from user prompts. The company said the model performed strongly on a broad benchmark test and can automatically select tools and draw on conversational context to respond more efficiently, keeping costs low.

Other players are also moving aggressively. Z.ai recently opened free access to a new version of its GLM model before temporarily restricting new sign-ups as demand overwhelmed its computing capacity. Meanwhile, shares of Baidu climbed to multi-year highs after it released an updated Ernie model, which the company said outperformed some competing systems on internal tests.

Industry observers say the rapid cadence reflects a broader strategy that goes beyond headline benchmark scores. Many Chinese AI models are open-sourced or offered at low cost, making them attractive to developers and governments in emerging markets. Analysts note that this approach could help Chinese firms build ecosystems and lock in usage, even if their most advanced models trail U.S. counterparts by a matter of months.

There are signs the strategy is gaining traction. Microsoft recently cited estimates suggesting DeepSeek’s adoption in parts of Africa is several times higher than in other regions. By lowering barriers to access and customization, Chinese developers hope to seed applications that rely on their underlying models.

At home, companies are also integrating AI more deeply into existing platforms. Tencent has rolled out promotions tied to its Yuanbao chatbot during the Lunar New Year, echoing tactics that once helped its WeChat app dominate digital payments. Alibaba, meanwhile, is linking its Qwen chatbot directly with Taobao, allowing users to shop, order food and pay without leaving the AI interface.

Analysts say this focus on traffic and integration may prove just as important as raw model performance. By embedding AI into popular commerce, gaming and social platforms, Chinese firms aim to offset high development costs and generate revenue through increased user engagement.

A year after DeepSeek’s debut raised questions about the effectiveness of U.S. tech controls, China’s AI sector appears firmly in acceleration mode. Rather than chasing a single “best” model, companies are racing to release, refine and distribute tools that can quickly scale—reshaping how and where global AI adoption takes place.