Meta pivots from open-source Llama to closed 'Avocado' frontier models

Reporting indicates Meta is developing frontier models code-named 'Avocado' that will debut in closed, proprietary form — a sharp departure from the open-weight Llama strategy that made Meta the standard-bearer for open AI. The effort is now led by Alexandr Wang, the Scale AI CEO who joined after Meta's roughly $14B investment in Scale, signaling a centralized, well-funded push toward a 'hybrid superintelligence' that is mostly closed.

The shift is strategically consequential. Llama's open releases seeded a vast ecosystem and pressured rivals; pulling back to closed models aligns Meta with OpenAI and Anthropic's playbook but risks alienating the developer community that championed it. That backlash is already visible — community threads lament the loss of free access and 'open-source democratization,' and Hugging Face CEO Clement Delangue posted that 'concentration of power, capabilities and economic wealth is the biggest risk in AI. We need open science and open-source more than ever!'

The move dovetails with Meta's other model news: Muse Spark, MSL's first publicly released model, now powers Meta AI on most of its smart glasses, replacing Llama 4 and narrowing the gap to leading systems. Together they suggest Meta is consolidating around a smaller number of higher-investment, more controlled models rather than broad open releases. The open question is whether Meta can match frontier rivals on raw capability while having ceded the open-ecosystem advantage it built — and whether the community migrates to Mistral, Qwen or DeepSeek in response.