Meta Platforms Inc. is reportedly steering its artificial intelligence efforts in a new direction with a next-generation model code-named “Avocado,” reflecting a significant departure from the company’s long-standing focus on open-source AI development. The model is expected to be launched in the spring of 2026 as a closed-source, proprietary system, a move that could reshape Meta’s role in the rapidly evolving AI landscape.
According to multiple industry reports, Meta’s internal AI teams are developing Avocado under tight controls, meaning that unlike previous models — such as those in the open-source Llama family — its core code and model weights will not be freely distributed to the public or developers. Instead, access is likely to be provided via APIs or hosted services, enabling Meta to monetise the technology more directly.
The Avocado project is being led by Meta’s TBD Lab within the AI Superintelligence Labs organisation, and the company has integrated technology from multiple third-party models, including inputs from Google’s Gemma and Alibaba’s Qwen, into its training pipeline. Meta is also reportedly investing heavily in infrastructure to support the model’s development, including a substantial order of high-performance chips for training clusters.
The shift away from open-source is widely seen as part of a broader strategic recalibration by CEO Mark Zuckerberg, who previously championed open-access AI as a distinguishing pillar of Meta’s strategy. Recent business reports suggest that the lukewarm reception to Llama 4 — particularly delayed deployments and mixed responses from developers — has prompted internal restructuring and a rethink of how Meta competes with rivals like OpenAI and Google in the frontier AI space.
Industry analysts view Avocado as a potential competitor to leading closed models such as OpenAI’s GPT series and Google’s Gemini, and its commercial orientation suggests Meta may leverage the model across its ecosystem — including upgrades to AI assistants embedded in Facebook, Instagram, and WhatsApp — while also offering enterprise API access to developers through cloud partners.
Despite the excitement around Avocado, timing and performance remain fluid. Multiple sources indicate that training and performance challenges have pushed the model’s debut to early 2026, and while Meta has not officially confirmed delays, internal reports describe a rigorous testing process designed to ensure robustness before launch.
The announcement of Avocado marks a notable pivot in Meta’s AI philosophy: from open-source democratization toward a proprietary, monetizable, and strategically controlled model. This change underscores the intensifying competition among tech giants to not only innovate but also to capture greater financial value from artificial intelligence.
