Why Meta’s Avocado AI Might Stay Closed Source
Meta’s cooking up a new AI model called Avocado, a potential successor to its Llama series. Unlike past Llama releases, this one could skip open source entirely and go proprietary. CNBC, Seeking Alpha, Mint, and Proactive Investors report insiders expect a debut in early 2026 after training tweaks, and people familiar say it might lock down weights and code from outsiders.
Lessons from Llama 4’s Stumble
The pivot kicked off with Llama 4’s April flop. Developers ignored it, prompting Mark Zuckerberg to rethink openness. In July, he wrote that open source needs caution to handle risks. WinBuzzer covers this as Meta abandoning open roots for closed models.
Rivals Ripping Off Llama
China’s DeepSeek R1 this year borrowed Llama architecture chunks, ticking off Meta staff. That incident spotlighted open source downsides—free tech gets copied fast, handing edges to competitors.
New Hires Push for Secrets
Meta dropped $14.3 billion in June on Scale AI founder Alexandr Wang and top talent, as StockTwits reports now running chief AI officer and TBD Lab where Avocado brews. Wang’s crew, plus Meta Superintelligence Labs leaders like Nat Friedman, question open strategies. They want a leading closed model to match OpenAI and Google. A Meta spokesperson insists open source stance holds, but actions point elsewhere.
Internal shakeups followed: Chris Cox lost AI oversight post-Llama 4. Layoffs hit, workweeks stretched to 70 hours. New bosses bring startup vibes, skipping Meta’s usual tools for quick demos over memos.
Meta’s ad machine prints money—over $160 billion yearly, up 20% from AI tweaks—but Zuckerberg eyes bigger AI wins. With stock lagging and capex jumping to $70-72 billion for 2025, Avocado’s a bet to catch up.
