Meta's Next-Gen AI Model 'Avocado' Reportedly Delayed After Lagging Behind Rivals in Key Tests
MENLO PARK, Calif. — Meta Platforms Inc. is facing a potential setback in its ambitious artificial intelligence roadmap. The company's much-anticipated foundational AI model, known internally by the codename "Avocado," has underperformed in recent internal benchmarks against leading competitors, prompting a likely delay to its public release, The New York Times reported Thursday, citing people with knowledge of the matter.
The evaluation reportedly placed "Avocado" behind the latest AI models from Alphabet Inc.'s Google, specifically its Gemini series, and from AI safety-focused startup Anthropic. While "Avocado" represents a significant leap over Meta's previous in-house models and even outperformed an earlier iteration of Google's Gemini, it has not kept pace with the most recent advancements from its rivals.
As a result, Meta has pushed back the launch timeline for "Avocado" from an initial target this month to at least May, the sources indicated. The model is intended to serve as the engine for a new generation of AI chatbots, coding assistants, and other integrated products across Meta's family of apps, including Facebook, Instagram, and WhatsApp.
The performance gap has led to internal discussions about temporarily licensing technology from Google's Gemini to bridge the capability shortfall for some products, though no agreement has been finalized. A Meta spokesperson did not immediately respond to a request for comment.
The reported delay underscores the intense pressure and high financial stakes in the generative AI race. CEO Mark Zuckerberg has positioned AI as the cornerstone of Meta's future, committing tens of billions of dollars to build out specialized data centers and hire top research talent. In a January earnings call, the company projected its total capital expenditures for the year could reach approximately $135 billion, nearly double the prior year's spend, largely to support its AI ambitions.
Beyond "Avocado," Meta continues development on other specialized models. A system focused on image and video generation, codenamed "Mango," is in the works, and future plans include a project under the moniker "Watermelon," according to the report.
Industry Reactions
Dr. Anya Sharma, AI Research Lead at Stanford University: "This is a common dynamic in fast-moving fields. Meta has strong research, but integration and scaling for a product used by billions is a different challenge than pure benchmark performance. A short delay for refinement is a prudent move."
Marcus Chen, CTO of a rival AI startup: "It highlights the difficulty of catching up once you fall behind in architecture and training scale. The open-source models from Meta have been influential, but competing at the very top tier with closed, commercially-tuned models requires immense, sustained execution."
David Park, tech analyst and frequent industry commentator: "This is embarrassing for Zuckerberg. After all the hype and billions spent, they're considering licensing from Google? It calls their whole 'AI-first' strategy into question. They're becoming a fast follower, not a leader."
Eliza Rodriguez, product manager at a major tech consultancy: "The delay itself is less critical than Meta's response. If they use this time to rigorously address safety and alignment issues that others may have rushed, it could become a strategic advantage, especially for consumer-facing applications."
Photo: PJ McDonnell / Shutterstock