Meta Delays Avocado AI Rollout as Gemini Licensing Talks Emerge
Meta has postponed the release of its frontier artificial intelligence model, codenamed Avocado, and now plans a launch no earlier than May. Previously, the company had targeted March for the model’s debut.
The delay adds another hurdle as Meta works to compete with major AI developers. Meanwhile, the company is reportedly considering licensing Google’s Gemini technology to strengthen its AI offerings while Avocado development continues.
Consequently, the potential partnership could help maintain momentum across Meta’s AI products until the new model becomes ready for release.
Repeated Timeline Shifts
Avocado has already faced several delays during development. Initially, insiders expected the model to arrive before the end of 2025. However, training and performance-testing challenges pushed the timeline into the first quarter of 2026.
Later, projections shifted again toward a possible spring launch. In addition, the company reportedly considered releasing the model as a closed, proprietary system rather than following its previous open-source strategy.
In early February, an internal memo from Meta Superintelligence Labs product manager Megan Fu described Avocado as “now Meta’s most capable pre-trained base model to date,” noting it was “competitive” with leading post-trained models even before undergoing post-training refinement.
Despite those internal assessments, the model has apparently struggled to match the performance levels set by competing AI systems. As a result, the development team continues to refine the model before releasing it publicly.
Mounting Pressure on AI Strategy
The delay highlights a broader period of uncertainty surrounding Meta’s AI efforts. Earlier flagship models also experienced setbacks, which forced the company to reassess its development strategy.
As a result, leadership introduced changes across its AI teams and adjusted its long-term approach to model development. At the same time, the company strengthened its collaboration with Google’s AI infrastructure.
Earlier this year, Meta signed a multi-billion-dollar agreement to lease Google’s Tensor Processing Units for training advanced AI systems. In addition, some internal proposals explored fine-tuning Gemini models using Meta’s advertising data to improve ad targeting.
Looking ahead, the company has forecast capital expenditures of at least $115 billion for 2026, much of which will support AI development. Therefore, Avocado’s performance against competing systems will likely play a critical role in evaluating the company’s recent AI investments.
