Key Points
- Meta has moved the launch date for its major AI model “Avocado” to at least May 2026.
- Performance testing places the model somewhere between Google’s Gemini 2.5 and Gemini 3.
- The company explores a potential short-term licensing agreement with Google for Gemini AI.
- Llama 3, released in 2024, remains Meta’s most recent frontier AI model.
- This postponement occurs alongside Meta’s planned capital expenditure of up to $135 billion in 2026.
Internal performance evaluations revealed that Meta’s “Avocado” AI model requires additional development time, leading to a postponement until May at the earliest. The company now explores licensing Google’s Gemini as an interim solution.
Meta’s Avocado AI Model Faces Development Challenges
The release timeline for Meta Platforms’ upcoming AI model, known internally as “Avocado,” has shifted from the current month to May — with June remaining a possibility.
The New York Times broke the story, drawing on information from people with direct knowledge of the situation. Reuters independently verified that benchmark testing shows the model performing at a level between Google’s Gemini 2.5 and Gemini 3 — below the company’s internal targets.
Llama 3, launched in 2024, stands as Meta’s most recent frontier-level AI model. Meanwhile, Google’s Gemini, OpenAI’s GPT series, and Anthropic’s Claude have continued advancing their capabilities.
Two streamlined versions of Llama 4 arrived last year, optimized for systems with constrained resources. A more robust model called “Behemoth” was expected to follow, though updates on that project have remained scarce.
A Meta spokesperson stated: “As we’ve said publicly, our next model will be good, but more importantly, show the rapid trajectory we’re on, and then we’ll steadily push the frontier over the course of the year as we continue to release new models.” The spokesperson added: “We’re excited for people to see what we’ve been cooking very soon.”
Meta shares declined approximately 2% during Friday’s early trading session.
Google Gemini Licensing Under Consideration
The situation takes an intriguing turn with reports that executives in Meta’s AI division have held discussions about securing temporary access to Google’s Gemini through a licensing arrangement. This would serve as a bridge solution while Avocado undergoes final development. The discussions remain preliminary, with no formal commitment established.
Such an arrangement would present a challenging moment for CEO Mark Zuckerberg. His Superintelligence Labs division has attracted leading AI researchers and engineers. Capital expenditure projections for 2026 range from $115 billion to $135 billion, with substantial portions allocated to AI infrastructure development.
A licensing deal would position Alphabet favorably in the competitive landscape — transforming its narrative from playing catch-up to becoming a sought-after AI provider that competitors seek to license.
Meta’s AI investments have already produced tangible benefits — enhanced advertising precision, refined content recommendation systems, and proprietary chips for recommendation algorithms. The company also commands a leading position in smart glasses technology.
Zuckerberg has articulated his vision for creating a “personal superintelligence” — a Meta-owned system capable of rivaling Apple and other smartphone platforms as the primary computing interface for daily tasks. Achieving this ambition requires proprietary AI leadership rather than dependence on licensed technology.
The Avocado postponement remains the central development. Meta shares traded down approximately 2% as of Friday’s early session.
