What’s next for Chinese open-source AI
The Rise of Chinese Open-Source AI: A Game-Changer for the Industry
The past year has marked a turning point for Chinese AI, with companies repeatedly delivering AI models that match the performance of leading Western models at a fraction of the cost. The latest open-weight model, Kimi K2.5, released by Moonshot AI, has come close to top proprietary systems such as Anthropic's Claude Opus on some early benchmarks. The difference: K2.5 is roughly one-seventh Opus's price.
China's Commitment to Open Source Will Continue
When DeepSeek launched R1, much of the initial shock centered on its origin. Suddenly, a Chinese team had released a reasoning model that could stand alongside the best systems from US labs. But the long tail of DeepSeek's impact had less to do with nationality than with distribution. R1 was released as an open-weight model under a permissive MIT license, allowing anyone to download, inspect, and deploy it. On top of that, DeepSeek also published a paper detailing its training process and techniques.
The Field Has Widened Rapidly
Since the success of DeepSeek, the field has widened rapidly. Companies such as Z.ai (formerly Zhipu), MiniMax, Tencent, and a growing number of smaller labs have released models that are competitive on reasoning, coding, and agent-style tasks. The growing number of capable models has sped up progress. Capabilities that once took months to make it to the open-source world now emerge within weeks, even days.
Open Source Has Become Politically Correct in China
In the Chinese programmer community, open source has become politically correct, framing it as a response to US dominance in proprietary AI systems. That shift is also reflected at the institutional level. Universities including Tsinghua have begun encouraging AI development and open-source contributions, while policymakers have moved to formalize those incentives.
The Next Wave of Models Will Be Narrower—and Better
Chinese open-source models are leading not just in download volume but also in variety. Alibaba's Qwen has become one of the most diversified open model families in circulation, offering a wide range of variants optimized for different uses. The lineup ranges from lightweight models that can run on a single laptop to large, multi-hundred-billion-parameter systems designed for data-center deployment.
The Open-Weight Nature of These Releases Makes It Easy for Others to Adapt Them
The open-weight nature of these releases also makes it easy for others to adapt them through techniques like fine-tuning and distillation, which means training a smaller model to mimic a larger one. According to ATOM (American Truly Open Models), a project by the AI researcher Nathan Lambert, by August 4, 2025, new model variations derived from Qwen were "more than 40%" of new Hugging Face language-model derivatives, while Llama had fallen to about 15%.
The Rise of Small Models
Compute and energy are real constraints for any deployment. The rise of small models is about making AI cheaper to run and easier for more people to use. His company, ModelBest, focuses on small language models designed to run locally on devices such as phones, cars, and other consumer hardware.
OpenClaw, an Open-Source AI Agent
OpenClaw, an open-source AI agent that recently went viral within the AI hacker world, allows AI to take over your computer—it can run 24-7, going through your emails and work tasks without supervision. OpenClaw, like many other open-source tools, allows users to connect to different AI models via an application programming interface, or API.
The Landscape of Open-Source Models in China Is Getting More Specialized
Research groups such as Shanghai AI Laboratory have released models geared toward scientific and technical tasks; several projects from Tencent have focused specifically on music generation. Ubiquant, a quantitative finance firm like DeepSeek's parent High-Flyer, has released an open model aimed at medical reasoning.
Chinese Open Models Will Become Infrastructure for Global AI Builders
The adoption of Chinese models is picking up in Silicon Valley, too. Martin Casado, a general partner at Andreessen Horowitz, has put a number on it: Among startups pitching with open-source stacks, there's about an 80% chance they're running on Chinese open models, according to a post he made on X. Usage data tells a similar story. OpenRouter, a middleman that tracks how people use different AI models through its API, shows Chinese open models rising from almost none in late 2024 to nearly 30% of usage in some recent weeks.
The Open-Source Ecosystems in China and the US Are Tightly Bound Together
The demand is also rising globally. Z.ai limited new subscriptions to its GLM coding plan (a coding tool based on its flagship GLM models) after demand surged, citing compute constraints. What's notable is where the demand is coming from: CNBC reports that the system's user base is primarily concentrated in the United States and China, followed by India, Japan, Brazil, and the UK.
The Interdependence of the Open-Source Ecosystems
That interdependence is part of what makes Chinese developers feel optimistic about this moment: The work travels, gets remixed, and actually shows up in products. But openness can also accelerate the competition. Dario Amodei, the CEO of Anthropic, made a version of this point after DeepSeek's 2025 releases: He wrote that export controls are "not a way to duck the competition" between the US and China, and that AI companies in the US "must have better models" if they want to prevail.
The Future of AI Development in China
For the past decade, the story of Chinese tech in the West has been one of big expectations that ran into scrutiny, restrictions, and political backlash. This time the export isn't just an app or a consumer platform. It's the underlying model layer that other people build on. Whether that will play out differently is still an open question.
Source: https://www.technologyreview.com/2026/02/12/1132811/whats-next-for-chinese-open-source-ai/
