Meta Delays Avocado AI Model to May After Internal Testing Reveals Performance Gaps with Google, OpenAI and Anthropic
Meta postpones its next-generation Avocado AI model by at least two months after benchmarks show it trailing rivals, raising questions about the company's $135 billion AI investment.
Overview
Meta has pushed the release of its next-generation AI model, codenamed Avocado, from March to at least May 2026 after internal testing revealed it trails the latest systems from Google, OpenAI, and Anthropic in key capability areas, according to The New York Times. The delay puts fresh scrutiny on the company’s plan to spend between $115 billion and $135 billion on AI infrastructure this year.
What We Know
Avocado was designed as Meta’s flagship frontier model and the successor to Llama 4, which itself disappointed developers after its April 2025 launch. Internal benchmarks show the new model outperforms Meta’s previous generations and surpasses Google’s Gemini 2.5, but it falls short of the newer Gemini 3.0, released in November, as well as the latest offerings from OpenAI and Anthropic, according to Reuters reporting via U.S. News. The specific areas of weakness include logical reasoning, programming, and writing.
A Meta spokesperson told reporters that “our next model will be good, but more importantly, show the rapid trajectory we’re on,” adding that the company is “excited for people to see what we’ve been cooking very soon.” An internal memo from a product manager at Meta Superintelligence Labs described Avocado as the company’s most capable pre-trained base model to date, suggesting it could outperform rivals after post-training improvements are applied.
Perhaps the most striking detail to emerge is that Meta’s AI leadership has discussed temporarily licensing Google’s Gemini technology to power certain Meta products while Avocado undergoes further refinement, according to Fortune. No decision has been confirmed, but the mere consideration represents a remarkable turn for a company that has long positioned its open-source Llama family as an alternative to proprietary AI systems, and a concession to a direct advertising rival.
What We Don’t Know
It remains unclear how close Meta came to actually approaching Google about a licensing arrangement, or whether such discussions moved beyond internal deliberation. The company has not disclosed specific benchmark results for Avocado, making independent performance assessments impossible. It is also uncertain whether the delay will extend beyond May, particularly given that Meta’s previous model, Llama 4, also experienced delays and underwhelming reception.
The financial implications of the postponement are similarly opaque. Meta has committed $115 billion to $135 billion in capital expenditure for 2026, a figure that matches the AI infrastructure spending of Amazon, Microsoft, and Google. Unlike those companies, however, Meta does not operate a cloud computing business that generates direct revenue from AI infrastructure. The return on that investment depends heavily on whether Meta can deploy competitive models across its platforms, including Facebook, Instagram, and WhatsApp.
Analysis
The Avocado delay marks the second consecutive stumble in Meta’s frontier model program. Llama 4’s lukewarm developer reception prompted a significant organizational overhaul, including the departure of chief AI scientist Yann LeCun to launch a startup, the hiring of Scale AI founder Alexandr Wang as chief AI officer in a deal valued at $14.3 billion, and an October restructuring that eliminated 600 positions across Meta Superintelligence Labs.
Those changes have yet to translate into a model that matches the performance of Meta’s primary competitors. Unlike Amazon, Microsoft, and Google, which can offset AI infrastructure costs through cloud computing revenue, Meta must justify its capital spending through improvements to advertising, messaging, and consumer products. With next-generation models codenamed Watermelon and Mango also in the pipeline, Meta’s AI roadmap faces a credibility test that the company’s record capital spending alone cannot resolve.