Meta’s AI Strategy Shifts Amid Delays to ‘Avocado’ Model
Key Points
- Meta’s Llama series pioneered open‑source large language models for developers.
- Meta AI, a consumer‑facing chatbot, expanded across Meta’s social platforms.
- The upcoming ‘Avocado’ model is designed as a proprietary AI system.
- Internal testing showed ‘Avocado’ lagging behind Google’s Gemini models.
- Launch of ‘Avocado’ has been delayed from March 2026 to May or June.
- Meta is considering a temporary license of Google’s Gemini to support its AI products.
- The shift toward closed‑source models aims to mitigate high AI development costs.
- Delays in both ‘Avocado’ and Llama 4’s ‘Behemoth’ raise concerns about Meta’s AI timeline.
Meta is reevaluating its artificial‑intelligence roadmap as the proprietary ‘Avocado’ model, slated for a 2026 release, encounters performance setbacks and timeline pushes. The company, once champion of open‑source Llama models, appears poised to move toward closed‑source solutions after internal tests showed ‘Avocado’ lagging behind rivals such as Google’s Gemini series. Discussions about temporarily licensing Gemini have surfaced, highlighting a potential reliance on external technology. These developments raise questions about Meta’s long‑term AI direction and its ability to stay competitive in a rapidly evolving market.
Meta’s AI Landscape
Meta has positioned itself as a major player in artificial intelligence through a blend of open‑source and proprietary initiatives. The Llama family, introduced as an open‑source multimodal large language model, was designed to democratize AI research and development. Over time, Meta released four Llama models and opened a limited preview of a Llama API, allowing developers to integrate the technology into their products.
In parallel, Meta launched Meta AI, a consumer‑facing chatbot integrated across its social platforms. Initially released as a feature within WhatsApp, Instagram, Facebook, and Messenger, Meta AI later received a standalone app with a Discover Feed, voice capabilities, and personalized content generation.
Introducing ‘Avocado’
Beyond Llama, Meta has been working on a next‑generation model code‑named ‘Avocado’. Unlike the open‑source Llama series, ‘Avocado’ is intended to be proprietary, meaning external developers would not have access to its weights or source code. A Meta spokesperson told Reuters that the model is meant to address performance gaps identified in internal testing.
Delays and Performance Concerns
Internal tests revealed that ‘Avocado’ fell short of competitors such as Google’s Gemini 2.5 and Gemini 3 in reasoning, coding, and writing tasks. Consequently, the launch originally expected in March 2026 has been pushed to May or June, according to a source familiar with the matter. The postponement mirrors earlier setbacks with Llama 4’s flagship model ‘Behemoth’, which has also been delayed as engineers work to improve its capabilities.
Strategic Implications
The shift from open‑source to a closed‑source model signals a broader strategic pivot for Meta. While open‑source models like Llama helped the company establish a foothold in the AI ecosystem, competitors such as DeepSeek have leveraged open components to build highly competitive systems. Moving to a proprietary model could help Meta offset the massive investments required for AI development, including a reported $14.3 billion stake in Scale AI to form Meta Superintelligence Labs, tasked with developing ‘Avocado’.
However, the possibility of temporarily licensing Google’s Gemini to power Meta’s AI products introduces a potential dependency on external technology, marking a departure from the company’s earlier goal of building core AI capabilities in‑house.
Outlook
Meta’s evolving AI strategy reflects tension between openness and control. The company must decide whether to continue investing in proprietary models like ‘Avocado’ or revert to open‑source approaches that foster broader adoption. As delays persist and performance gaps remain, Meta’s ability to maintain a leading position in the AI race will depend on clarifying its long‑term vision and delivering competitive technology.