Why Meta Pushed Back the Avocado A.I. Model

Meta has delayed the public rollout of its next-generation A.I. model, codenamed Avocado, after internal performance and training checks revealed shortcomings that require further work. An internal memo dated January 20 described Avocado as Meta’s most capable pre-trained base model to date, yet the model will undergo additional fine-tuning, safety evaluations, and scalability tests before it is broadly distributed. The company has substantially increased capital spending on A.I. infrastructure for 2026, with projected expenditures expected to rise into the triple-digit billions, and these resources will be applied to additional training runs and engineering improvements.

The development path for Avocado also marks a strategic shift. Where earlier Llama models were widely shared, Avocado is being prepared as a more tightly controlled system, which may limit early access and channel distribution through paid or partner-only offerings. Early engineering notes indicate emphasis on code-writing, multilingual ability, and vision-language understanding, but the work to stabilize long-context reasoning and video understanding remains ongoing.

Engineers will continue iterative evaluation, and limited beta testing is likely to precede any commercial launch in the first quarter of 2026. Investors, enterprise customers, and developer communities will watch how further testing changes both the model’s capabilities and Meta’s release plans. Nevertheless, the delay gives the company time to close performance gaps, and the eventual rollout will show whether the added compute and engineering effort produces a meaningful leap over prior models. If successful, Avocado could be positioned for enterprise products and paid APIs, expanding Meta’s AI offerings.