Meta is preparing a major shift in its artificial intelligence roadmap, and at the center of that change is a new model codenamed Avocado. Early reports suggest this system marks a significant departure from the company’s earlier approach that relied on releasing open models like the Llama family. With Avocado, Meta appears ready to prioritise commercial returns over openness, signalling a new phase in its AI ambitions.
For years, Meta positioned itself as the tech giant most committed to open AI development. Llama 2 and Llama 3 were shared freely with developers, research labs, and enterprises, giving Meta broad influence despite limitations in direct monetisation. Internally, though, executives began questioning how far openness could carry the company in a market dominated by paid platforms from OpenAI, Google, and Anthropic. The verdict was clear. The next wave of models needed to support revenue generation, not just ecosystem goodwill. Avocado is the first model created under this revised strategy and is expected to be available through paid APIs rather than free or open releases.
Also read: World’s 1st LLM trained in space: From earth to orbit on NVIDIA H100 GPUs
The development of Avocado coincides with a major restructuring of Meta’s AI efforts. The company has consolidated research groups into a new unit called Meta Superintelligence Labs, led by Alexandr Wang, founder of Scale AI. His leadership represents a more competitive, product-focused direction. Rather than broad academic-style work, teams are now concentrating on delivering frontier-class models that can stand alongside GPT, Claude, and Gemini. This reorganised structure has also aligned resources toward high-performance training runs and monetisable applications, reducing emphasis on long-term exploratory research.
What sets Avocado apart from Meta’s earlier projects is not just its capabilities but its likely release strategy. The model may not include downloadable weights or permissive licences. Instead, it could be gated behind enterprise agreements, a move that brings Meta into a more direct commercial rivalry with other AI leaders. This shift has triggered debate inside the company, especially among researchers who previously advocated for open access. Some high-profile departures from Meta’s AI teams in recent months have added to the perception of an internal philosophical split.
Also read: ChatGPT with Photoshop and Acrobat lowers Adobe’s learning curve, here’s how
For the broader AI ecosystem, Avocado represents a turning point. If Meta steps away from open releases, the landscape of accessible, state-of-the-art models may narrow. Standards groups and developers have already begun discussing the implications of fewer open alternatives. Meta has not announced a release timeline, but Avocado clearly signals its intent to compete at the top tier of AI while finally attaching a business model to its ambitions.
