Meta’s New Avocado AI Model Nears Launch Next Year
A new AI model, codenamed Avocado, may arrive early next year as part of work from the recently formed Meta Superintelligence Labs. This unit emerged after its leadership grew dissatisfied with results from current Llama models and aimed to move toward superintelligence, a stage where AI handles most tasks at human levels. Earlier this year, the company drew talent from major AI players to push this ambitious goal. Moreover, internal momentum appears strong, and leadership has repeatedly hinted that future releases could surpass industry standards.
Although no official updates have surfaced about ongoing development, the company’s chief executive expressed confidence in July that upcoming Llama versions would be “the most advanced in the industry” and would “bring the benefits of AI to everyone.” Yet, despite similar progress from rivals like Google, OpenAI, and Anthropic, no new Llama models have appeared. However, a recent report suggests Avocado is now the primary focus.
Delayed Timeline and Development Challenges
Initially, many expected Avocado to launch before the end of this year. However, the schedule now points to the first quarter of 2026. Even so, this shift appears linked to training and performance-testing challenges that the team must resolve to ensure a strong debut. Although the timeline changed, a spokesperson noted, “Our model training efforts are going according to plan and have had no meaningful timing changes.”
Still, the broader race for AI advancement continues, and the delay adds pressure as competitors move quickly. Nevertheless, internal teams seem confident, and the extended window may help deliver a more refined model.
Possible Move Away From Open Source
While leadership previously advocated strongly for open-source AI, the stance has begun to shift. In July, the company’s chief said, “We’ll need to be rigorous about mitigating these risks and careful about what we choose to open source.” Consequently, the new Avocado model may follow a different path. Reports indicate it could be proprietary, similar to Gemini and ChatGPT, preventing developers from freely downloading model weights or examining key components.
Even with this potential shift, leadership remains upbeat about the long-term vision. During an earnings call earlier this year, the chief executive said, “I think that we’ve already built the lab with the highest talent density in the industry… We’re heads down developing our next generation of models and products, and I’m looking forward to sharing more on that front over the coming months.”
