Chinese AI models are making their way to the top ends of benchmarks and leaderboards, and VCs are already seeing the difference on the ground.
Martin Casado, a partner at a16z, has said that a majority of startups that pitch to them now use Chinese AI models. “I’d say 80% chance (they are) using a Chinese open-source model,” he says in an interview with The Economist.
Chinese AI models have grown in prominence in recent times. While DeepSeek was the first Chinese AI model that had gone mainstream, there are now dozes of Chinese companies that are making extremely capable AI models. These models can generate text, images, and videos, and are often much cheaper than models released by top US-based labs. More crucially, these models are open-weights, which means that companies can tweak them and host them independently without being dependent on their creators or other third parties.
And China’s dominance in this space has only increased in recent times. A Chinese model, Kimi K2 Thinking, is currently placed ahead of the best closed models from Google, Anthropic and Grok on Artificial Analysis’ Intelligence Leaderboard, and is only behind OpenAI’s GPT-5. The model is also open-weights, which means that it is the best open model in the world. Apart from Kimi K2, Chinese models like MiniMax-M2, Qwen3, DeepSeek and GLM are all in the top 15, providing cheaper and open alternatives to the best American models. a16z’s Marc Andreessen has often emphasized on how the US needs to beat China at AI, but with 80% of American startups that his firms see already using Chinese AI models, the US would need to take urgent steps to make sure that it isn’t left behind in the AI race.