China's open-source models lead: Alibaba's Qianwen surpasses Meta's Llama
Alibaba's open-source model Qwen has surpassed Meta's Llama in global download volume and the number of derivative models, becoming the world's leading open-source model. The new version Qwen3 performs at the level of GPT-5 and Gemini 2.5-Pro, supporting 119 languages. Minimax's M2 and Kimi's K2 have also received global acclaim for their powerful capabilities and low costs. Silicon Valley investor Chamath Palihapitiya revealed that his company has migrated a large workload to Kimi K2, stating that its performance exceeds that of OpenAI and Anthropic while being more cost-effective
Qwen's global download volume and the number of derivative models have surpassed the Meta Llama series from the United States, ranking it as the world's number one open-source model; the newly updated Qwen3's performance has reached the level of GPT-5 and Gemini 2.5-Pro, supporting 119 languages.
In addition to Alibaba's Qwen, Minimax's M2 and Kimi's K2 are also highly praised globally for their powerful capabilities and low costs.
Even Silicon Valley is quietly running China's open-source models.
Venture capitalist Chamath Palihapitiya revealed in a podcast last month that his venture capital firm has migrated a significant amount of workload to the open-source model Kimi K2 on the dark side of the moon.
He stated that Kimi K2 "performs much better and, frankly, is much cheaper than OpenAI and Anthropic."
