Chinese LLMs claim six spots in top 10 models on OpenRouter
Six Chinese large language models rank in the top 10 on OpenRouter's LLM leaderboard this week, according to data from the popular AI gateway for developers. The models include Step 3.5 Flash, MiniMax M2.5, and DeepSeek V3.2.
On OpenRouter's market share ranking by model author, Xiaomi leads with 21.4 percent of token usage, followed by Google at 12 percent.
Cursor's coding tool relies on Chinese open-source model
Code-generation startup Cursor disclosed this week that its Composer 2 model uses Kimi K2.5, an open-source model from Chinese company Moonshot, as its foundation. The disclosure came after a developer spotted references to the Kimi model in Cursor's system.
Cursor founder Aman Sanger said on X that Kimi K2.5 "proved to be the strongest" among the models his team evaluated. He added that the company should have credited the Kimi base model in its initial announcement.
Moonshot's Kimi.ai account confirmed the partnership on X, describing it as an authorized commercial arrangement. The account said Cursor's team applied "continued pretraining & high-compute RL training" to the base model.
Performance and cost advantage drive adoption
Chinese LLMs are gaining traction among developers because they deliver strong performance at lower inference costs than competitors, according to Tian Feng, former dean of SenseTime's Intelligence Industry Research Institute.
Tian said the Cursor-Kimi partnership demonstrates that Chinese models are "continuously driving global AI development." He noted that open-source cooperation among Chinese LLM developers has moved beyond strategy into market validation.
The shift reflects how developers choose models based on benchmarks and real-world performance rather than vendor origin.
Your membership also unlocks:
