Here’s a brief update on a positive development for domestic AI models. On the evening of December 10, Alibaba's U.S. pre-market shares surged sharply, maintaining a gain of around 2% after the official market opening.
The news stems from Meta CEO Mark Zuckerberg, who, months after assembling one of the most expensive tech teams in history, has personally stepped in to oversee daily operations. He is steering the company toward an AI model that can actually generate revenue.
According to insiders, a new model codenamed "Avocado" is expected to debut next spring and will likely be released as a "closed-source" model—one that can be tightly controlled, with Meta selling usage rights. This approach aligns with competitors like Google and OpenAI and marks the biggest departure yet from Meta’s long-standing open-source advocacy. Open-source models allow external developers and researchers to review and build upon the code. Sources say Meta’s newly appointed Chief AI Officer, Alexandr Wang, favors closed-source models.
Earlier this year, Meta’s strategy took a major turn after the release of its open-source model Llama 4, which disappointed both Silicon Valley and Zuckerberg. Zuckerberg sidelined some team members involved in the project while personally recruiting top AI researchers and leaders—some with multi-year compensation packages worth hundreds of millions of dollars. Now, Zuckerberg is dedicating significant time and effort to collaborating closely with these new hires, who have been consolidated into a team called TBD Lab.
Insiders reveal that TBD Lab is using multiple third-party models to "distill" Avocado’s training, including competitors like Google’s Gemma, OpenAI’s GPT-OSS, and Alibaba Group’s Qwen (Tongyi Qianwen).
The use of Chinese technology to train the new model signals a subtle shift in Zuckerberg’s stance. Llama and other U.S. efforts have begun to lag. As Nvidia CEO Jensen Huang noted earlier this month, "China is far ahead in open-source—way ahead."
Comments