You are currently viewing Meta’s Avocado Model: Training Delays and a Pivot from Llama
Featured image for Meta's Avocado Model: Training Delays and a Pivot from Llama

Meta’s Avocado Model: Training Delays and a Pivot from Llama

  • Post author:
  • Post category:News
  • Post comments:0 Comments

Meta’s Avocado Model: Training Delays and a Pivot from Llama

Image sourced from cnbc.com
Image sourced from cnbc.com

Meta’s developing Avocado, a new frontier AI model meant to follow Llama. Insiders told CNBC that many expected a release by end of 2025, but it’s slipped to Q1 2026. The holdup? Training performance tests to ensure it performs well on debut. A Meta spokesperson pushed back: “Our model training efforts are going according to plan and have had no meaningful timing changes.”

Avocado comes from TBD Lab, led by new chief AI officer Alexandr Wang. Meta shelled out $14.3 billion in June for Wang, Scale AI’s founder, plus top engineers and researchers. That buy included a stake in Scale. Wang heads the lab near Zuckerberg’s office; the team skips Meta’s internal Workplace network and runs like its own startup.

Proprietary Shift?

Unlike Llama’s open-source roots, Avocado might stay closed off. Developers wouldn’t get free access to its weights. The Verge points out Meta could charge for access to future models like this. Zuckerberg hinted at caution on open source after Llama 4 bombed with developers in April. DeepSeek’s R1 using Llama bits upset some at Meta.

New Hires Reshape AI Teams

Meta grabbed more talent: former GitHub CEO Nat Friedman now leads product and research in Meta Superintelligence Labs (MSL), and Shengjia Zhao, ChatGPT co-creator, joined too. They brought Silicon Valley methods that clash with Meta’s old software process. Friedman pushes “Demo, don’t memo” and new tools for AI agents.

  • Chris Cox, longtime chief product officer, lost oversight of the GenAI unit after Llama 4.
  • 70-hour weeks are common; layoffs hit MSL in October, contributing to Yann LeCun leaving for a startup.
  • Vishal Shah moved to VP of AI Products; Aparna Ramani handles MSL compute resources.

Pressure mounts from rivals. Google dropped Gemini 3 last month. OpenAI updated GPT-5; Anthropic launched Claude Opus 4.5 in November. Nvidia’s Jensen Huang name-checked most in an earnings call—Meta’s Llama got no mention.

Meta raised 2025 capex to $70-72 billion. Analysts at KeyBanc say Meta’s flipped from AI leader to facing investor scrutiny, unlike Alphabet. Still, Zuckerberg calls MSL the highest talent density around.

More stories at letsjustdoai.com

Seb

I love AI and automations, I enjoy seeing how it can make my life easier. I have a background in computational sciences and worked in academia, industry and as consultant. This is my journey about how I learn and use AI.

Leave a Reply