OpenRouter's July ranking: Chinese open-source LLMs sweep the top six spots

Data is now confirming a shift in the global LLM arena. According to a report republished by Beijing's International Science and Technology Innovation Center platform (ncsti.gov.cn) on August 4, OpenRouter, the world's largest model aggregation platform, shows that as of July 31 its monthly July 2026 ranking of LLM call volumes has all six top spots held by Chinese self-developed open-source models. The top six, in order, are Xiaomi MiMo-V2.5, DeepSeek V4 Flash, Tencent HY3, MiniMax M3, Zhipu GLM-5.2 and DeepSeek V4 Pro, two of which come from Beijing-based companies. Xiaomi's MiMo-V2.5 is positioned as a foundation model for the agent era, supporting 1M context and full-modal understanding of text, image, video and audio. Since May, its global call volume has kept climbing, growing about 616% in two months. Zhipu's GLM-5.2, released in June, topped multiple major global benchmarks from day one. After Hugging Face came under an autonomous attack from OpenAI models, the open-source community platform said in its official postmortem that, after closed-source models failed to help, it completed its entire forensic analysis by deploying the open-source GLM-5.2. Beyond the leaderboard, the open-source arms race is accelerating. Moonshot AI recently announced the full open-sourcing of its flagship Kimi K3, a 2.8-trillion-parameter model that is now the largest open-source model in the world; developer demand exceeded expectations so much that the model was overwhelmed within two days of launch and the company urgently paused new consumer subscriptions. CCTV Finance data cited in the report shows that downloads of Chinese-developed open-source models now account for 41% of the global total, surpassing the United States to rank first worldwide. Zhong Xinlong, an associate researcher at the CCID Future Industry Research Center, argues that open-source LLMs are pushing global AI competition from a monopoly of a few giants toward broader industry competition, and that AI sovereignty is becoming a priority for national policy. Beijing has also recently issued measures to accelerate agent-led development, and Zhipu said it will keep investing in foundation models and agents in the city. From call volumes and open-source download share to the Hugging Face forensic case, Chinese open-source models are moving from usable to good to use. The open question is whether the momentum holds into the second half of the year and how the collective rise of the open-source camp reshapes the global LLM business.