Once a staunch advocate of open source and proudly called itself the "Android of the AI era," Meta is now quietly taking a completely different path. According to a recent Bloomberg report, Meta plans to release a new AI model codenamed "Avocado" (Guac) in the spring of 2026—and this model is likely to be closed-source. Even more shocking is that it reportedly used Alibaba's open-source large model Qwen directly during its training process.
This news instantly sparked the market. Alibaba's stock surged as much as 4% before the U.S. market opened and finally closed up 2.53%. Netizens joked: "Silicon Valley closed-source models are actually trained on Chinese open-source models." Irony lies in the fact that Meta, once a tech giant that held open source as a belief, now relies on the achievements of global open-source communities—including Alibaba's Qwen, Google's Gemma, and OpenAI's gpt-oss—to build its own closed-source products.
This is not just a shift in technical strategy but also a collapse of strategic beliefs. In the past, Zuckerberg believed that open source could accelerate AI adoption and expand ecological influence, even comparing the Llama series to the "Android system of the AI era." However, the lackluster release of Llama4 in April 2025 became a turning point. The model was exposed to suspicion of "cheating in rankings" in the authoritative LMArena evaluation, leading to a sharp decline in reputation. In anger, Zuckerberg quickly marginalized the original Llama team and personally launched a "dream team" recruitment plan—offering multi-million-dollar salaries and even reports of "bringing homemade pumpkin soup" to persuade core researchers from OpenAI.
In this major reshuffle, key figures have emerged: Alexander Wang, Meta's new Chief AI Officer and a firm supporter of the closed-source approach. He came from Scale AI with a $1.43 billion deal and quickly took control of the AI strategy. His TBD Lab became the company's new core, with an office even located next to Zuckerberg himself, who almost daily checked the progress and called himself "Wang's mentor."
However, this excessive attention seems to have had the opposite effect. According to insiders, Alexander Wang has already shown impatience with his "personal mentor"—no matter how much power one has, it's hard to bear a boss constantly pacing near your desk.
With the strategic shift, Meta's open-source foundation has been systematically dismantled. Starting in October 2025, the FAIR lab laid off a large number of staff, including research scientist director Tian Yuan Dong, who was also not spared. In November, Yann LeCun, known as the "father of convolutional neural networks," left the company quietly, with the reason directly pointing to the company no longer supporting his open-source research direction. There were rumors that before leaving, he was even asked to "reduce public appearances" to avoid interfering with the new strategic narrative.
Today, Meta's AI future is firmly tied to "Avocado." Yet ironically, this closed-source model, which aims to end the open-source era, is deeply rooted in open-source soil—especially Qwen, created by Chinese teams. When Silicon Valley giants start "copying homework" and try to privatize the results, the market finally sees clearly: true technological authority may have long since moved away from Mountain View.
Meta's "elephant turn" is not a victory song, but a requiem for open-source idealism. And in this global AI competition, that once underestimated Eastern open-source force is quietly becoming the cornerstone of others' closed-source ambitions.
