From ‘Llama’ to ‘Avocado’: Meta’s Sudden AI Shift Has Developers Worried

Remember when Mark Zuckerberg was all about giving away Meta’s AI secrets? Those days appear to be over.

While Meta has spent the last year hyping up its “Llama” family of AI models, the company is quietly pivoting to a new project codenamed “Avocado.” However, reports suggest that this new flagship model is facing delays and a major philosophical shift that could change how the world accesses Meta’s technology.

For a while, the expectation inside Meta was that their next big thing would arrive before we ring in the new year. That timeline appears to have slipped.

According to a CNBC report, the company’s new frontier model, Avocado, is now targeted for release in the first quarter of 2026. Sources familiar with the matter told the outlet that the model is currently “wrestling with various training-related performance testing intended to ensure the system is well received when it eventually debuts.”

Despite the internal chatter about delays, the company is officially putting on a brave face. In that same report, a spokesperson for Meta told CNBC: “Our model training efforts are going according to plan and have had no meaningful timing changes.”

Closing the doors?

The biggest news isn’t the name or the delay, but a potential U-turn in strategy. Until recently, Meta was the hero of the open-source community, giving away its Llama technology so developers worldwide could peek under the hood and use it freely.

That era might be ending. CNBC reports that Avocado could launch as a proprietary, closed model. This means outside developers wouldn’t be able to download the “weights” or software components as they did with Llama.

This shift seems to be driven by security concerns and stiff competition. There was reportedly internal frustration after foreign competitors, including Chinese AI labs, used Llama’s open architecture to build their own rival tools.

Mark Zuckerberg himself hinted at this change over the summer. “We’ll need to be rigorous about mitigating these risks and careful about what we choose to open source,” said Zuckerberg in July, according to CNBC.

A rebuilt AI leadership

Behind the strategic turn is a leadership overhaul. After Llama 4’s underwhelming launch, Meta reorganized parts of its AI division and made one of the industry’s most expensive talent bets: a $14.3 billion acquisition of Scale AI founder Alexandr Wang and several top engineers.

Wang now leads a new elite group called TBD Lab, where Avocado is being built. Meta also recruited high-profile names like former GitHub CEO Nat Friedman and Shengjia Zhao, a co-creator of ChatGPT. CNBC reports that these leaders have pushed for faster development cycles and more modern tooling, clashing at times with Meta’s older, more consensus-driven engineering culture.

On the company’s October earnings call, Zuckerberg defended the reshaped team, saying, “I think that we’ve already built the lab with the highest talent density in the industry. We’re heads down developing our next generation of models and products and I’m looking forward to sharing more on that front over the coming months.”

Pressure from all sides

Meta enters 2026 under pressure on several fronts. Its competitors — OpenAI, Google, and Anthropic — are releasing upgraded models at a pace that leaves little room for missteps. Google’s Gemini 3 and OpenAI’s latest GPT-5 updates have been particularly well received. Within Meta, CNBC reports that 70-hour workweeks and multiple reorganizations have become common as teams scramble to keep the company competitive.

Meanwhile, Wall Street is watching closely. Meta’s stock has lagged Alphabet this year, and the firm has raised its 2025 capital expenditure forecast to between $70 billion and $72 billion. KeyBanc analysts wrote in November that “Meta has been the opposite of Alphabet,” entering the year as an AI winner but ending it with “more questions around investment levels and ROI,” according to the CNBC report.

Some parts of Meta’s business remain solid. Its advertising engine, powered significantly by AI, continues to grow at more than 20% a year. But Zuckerberg’s ambitions extend far beyond ads, and the company is now betting heavily on a high-end, closed AI model that can compete directly with Gemini, ChatGPT, Claude, and others.

For a look at which companies are actually pulling ahead with generative AI, this report on OpenAI’s State of Enterprise AI report breaks down adoption gaps, productivity gains, and how frontier firms are widening the lead.