Meta is reportedly developing its next-generation flagship large model, codenamed Avocado, with the target release date postponed from the original end of 2025 to the first quarter of 2026. Multiple insiders stated that the model will be positioned as the "successor" to the Llama series and may abandon the open-source tradition, adopting a closed-source commercial approach to directly compete with the closed-model ecosystems of OpenAI, Google, and Anthropic.

Model Positioning: Llama Successor + Closed-Source Commercialization  

- Function Level: Avocado is described as a "frontier-level" large model, with performance goals targeting GPT-5 and Gemini 3 Ultra  

- Open-Source Strategy: Internal discussions are evaluating a complete closed-source approach, offering only through API and hosted services, no longer providing weight downloads  

- Technical Approach: Integrating Llama architecture improvements + a new multimodal encoder, focusing on optimizing long context, tool calls, and reasoning speed

Timeline: Internal Delivery by Year-End → Release in Q1 2026  

- Originally planned for internal distribution by December 2025, it was postponed to Q1 2026 due to unmet performance testing and training optimization standards  

- Meta Spokesperson Response: "Model training is proceeding as planned, with no major changes to the schedule"

 Resource Investment: 1.5 billion dollars to attract talent + 2.7 billion dollars to build data centers  

- Human Resources: In June 2025, Meta acquired 1.43 billion dollars worth of shares in Scale AI and appointed its CEO Alexandr Wang as Meta's Chief AI Officer; top talents such as former GitHub CEO Nat Friedman and co-creator of ChatGPT Zhao Shengjia have joined successively  

- Computing Power: A 2.7 billion dollar joint venture with Blue Owl Capital to build the Hyperion Data Center in Louisiana, specifically serving Avocado training and inference

Strategic Shift: From "Open-Source Giving" to "Closed-Source Monetization"  

- Llama4 Failure: The release of Llama4 in April 2025 did not meet developer expectations, with internal admission that "performance and popularity were behind competitors," becoming a catalyst for Zuckerberg's strategic shift  

- Competitive Pressure: OpenAI GPT-5, Google Gemini 3 Ultra, and Anthropic Claude 3.5 have already been closed and offer enterprise APIs; Meta decided to "compete with closed models using closed models"  

- Revenue Model: Expected to monetize through API calls, enterprise hosting, and advertising-enhanced Meta AI, no longer relying on open-source community donations

Market Impact: Another Player Enters the Closed-Model Arena