CSIS says Chinese AI models are narrowing the gap with US systems
New releases suggest DeepSeek was part of a broader Chinese AI catch-up pattern.
Chinese AI models are narrowing the gap with leading US systems, according to a new analysis by the Center for Strategic and International Studies.
CSIS said recent releases from Z.ai, Moonshot, DeepSeek and Alibaba-backed Qwen show that China’s rapid progress in AI was not limited to DeepSeek-R1, but reflects a broader pattern of fast technical catch-up.
The analysis points to Z.ai’s GLM-5.2 model, which performs close to the top US closed models in coding and agent-based tasks. It also highlights strong results from Moonshot’s Kimi, DeepSeek V4-Pro and Qwen3.7-Max across software engineering, reasoning and agent benchmarks.
CSIS argues that Chinese models are now only months, rather than years, behind US frontier systems in several practical areas.
The report identifies knowledge distillation, open-weight research communities and efficiency-driven engineering as key factors behind this progress. Chinese labs can learn quickly from stronger models, shared research practices and open-source ecosystems, while US chip export controls have pushed them towards more efficient training and inference strategies.
Cost is another important factor. CSIS said Chinese models are often cheaper to access than leading US closed systems because open-source releases can be hosted by many providers, increasing price competition and making them easier for developers and governments to adopt.
The analysis says US firms still retain major advantages in frontier capabilities, cloud platforms, enterprise products and user feedback loops. However, Chinese models are now capable, affordable and open enough to shape global AI competition.
CSIS argues that US policy should therefore focus not only on protecting technological advantage, but also on building global trust, lowering access costs and ensuring partners see the American AI stack as reliable.
Why does it matter?
The analysis shows that AI competition is not only about which country has the most powerful frontier model. Chinese open-weight models are spreading because they are increasingly capable, cheaper to run and easier to deploy through third-party hosts or local infrastructure. That could shape global adoption, especially for governments, startups and developers that cannot afford or do not want to depend entirely on US closed-model providers. For the US, the challenge is no longer only maintaining a technical lead, but also making its AI ecosystem trusted, affordable and reliable for international partners.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!