Meta Delays AI Model ‘Avocado’ to Mid-2026
Meta’s next-gen AI model, Avocado, has been delayed until mid-2026 after internal tests showed it lagged behind OpenAI and Google in performance. This setback signals that even tech giants struggle to keep pace in agentic and generative AI, impacting post-production teams relying on timely transcription and multimodal AI improvements reported.
Meta's Avocado delay extends a pattern of AI setbacks, following previous issues with Llama 4 and Behemoth models that failed to meet internal expectations on reasoning and math benchmarks. Avocado's performance now ranks between Google’s Gemini 2.5 and Gemini 3.0, falling short of Meta’s goal to produce a frontier-class model this year. This has pushed the release from March to at least May or June 2026. The company is considering an unusual stopgap: temporarily licensing Google’s Gemini model to power some AI products, including its assistant across WhatsApp, Instagram, and Facebook. This potential move underscores the urgency for Meta to remain competitive amid rapid advances by Google, OpenAI, and Anthropic, despite the rivalry between the firms. Avocado marks a strategic shift for Meta, moving away from its long-standing open-source approach with Llama models toward a proprietary, closed-source system aimed at increasing profit margins and controlling distribution. This change reflects Meta’s effort to build a stronger economic moat and commercial footing in the AI landscape. Meta’s AI investment commitments remain massive, with planned capital expenditures between $115 billion and $135 billion in 2026, nearly doubling the prior year’s spend. This includes investments in data centers, custom chip development, and the new Meta Superintelligence Labs led by Alexandr Wang, who joined after Meta’s $14.3 billion acquisition of Scale AI. The delay affects post-production workflows relying on AI transcription and multimodal capabilities, as Avocado was expected to enhance these tools’ speed and accuracy. Meta’s struggle to field a competitive model this year may slow AI-driven innovations in video editing, color correction, and automated content tagging that consultants and studios eagerly anticipate.
