Podcast Episode
The rise has been swift. Chinese open-source models grew from just 1.2 percent of global usage in late 2024 to nearly 30 percent by late 2025, according to data from OpenRouter and Andreessen Horowitz. Among AI startups using open-source models, an estimated 80 percent now rely on Chinese offerings.
The competitive pressure comes as Meta has reportedly shifted toward closed-source development with its new Avocado model, abandoning its open-source Llama tradition following disappointing results from Llama 4. Google DeepMind CEO Demis Hassabis has acknowledged that Chinese AI models may be just months behind those developed in the United States.
Chinese AI Labs Sweep All Top Spots on Open-Source Leaderboard as US Scrambles to Respond
February 3, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Chinese artificial intelligence labs now hold all six top positions on independent open-source AI leaderboards, capturing nearly thirty percent of global AI usage. The dominance has triggered a funding race among American startups seeking to develop domestic alternatives, while Meta retreats from open-source development entirely.
China Takes the Crown in Open-Source AI
In a remarkable shift in the global artificial intelligence landscape, Chinese labs now claim all six top spots on independent open-source AI benchmark leaderboards, according to Artificial Analysis. The leading models include Moonshot AI's Kimi K2.5, Zhipu AI's GLM-4.7, and DeepSeek V3.2, with Kimi K2.5 and GLM-4.7 tied as the highest-intelligence open-source models available.The rise has been swift. Chinese open-source models grew from just 1.2 percent of global usage in late 2024 to nearly 30 percent by late 2025, according to data from OpenRouter and Andreessen Horowitz. Among AI startups using open-source models, an estimated 80 percent now rely on Chinese offerings.
Pre-Lunar New Year Launch Frenzy
Chinese AI labs are racing to release major updates before Lunar New Year festivities begin the week of February 15th. Zhipu AI plans to launch GLM-5 within two weeks, MiniMax is preparing M2.2 with coding enhancements, and Baidu has already unveiled Ernie 5.0, a 2.4 trillion parameter omni-modal model ranking first among Chinese models on the LMArena leaderboard. Alibaba released Qwen3-Max-Thinking, while tech giants Tencent, Baidu, and Alibaba are deploying billion-yuan giveaway campaigns to drive chatbot adoption.America's Open-Source Response
The Chinese dominance has prompted urgent action from American startups. San Francisco-based Arcee AI is seeking over 200 million dollars in funding at a valuation above one billion dollars, having recently launched Trinity Large, a 400 billion parameter foundation model trained for just 20 million dollars. Meanwhile, Reflection AI secured 2 billion dollars last year from investors including Nvidia to develop top-tier American open models.The competitive pressure comes as Meta has reportedly shifted toward closed-source development with its new Avocado model, abandoning its open-source Llama tradition following disappointing results from Llama 4. Google DeepMind CEO Demis Hassabis has acknowledged that Chinese AI models may be just months behind those developed in the United States.
Published February 3, 2026 at 6:25am
