Meta AI Could Reportedly Get New ‘Avocado’ Models, AI Agents and OpenClaw Integration

Unlike ChatGPT, Gemini, Claude, and other major artificial intelligence (AI) platforms, Meta AI been relatively slow in shipping new features and products. However, a new report claims that the Mark Zuckerberg-owned tech giant is working on several major capabilities to improve the overall experience of the app. Some of the features are aimed at increasing interoperability via integration of Anthropic's Model Context Protocol (MCP) and AI agents, and others are designed to improve the overall functionality of the Meta AI platform.

Meta AI to Reportedly Get New Features

According to a TestingCatalog report, Meta is planning to add several features to its AI chatbot. The publication found evidence of these features within the codebase of the Meta AI website and was even able to surface a few of them. However, since all of these are under development, they are not accessible to users.

The publication claimed to have found multiple internal modes in the platform apart from Fast and Thinking. Within these modes, traces of two AI models, dubbed Avocado and Avocado Thinking, were also spotted. Not a lot is known about these models or modes. However, the publication claimed that using Avocado (the only working model) to generate responses did not give any improved quality. However, this could be due to the server rerouting the queries via existing models.

In December 2025, Meta acquired Manus AI, a company known for its autonomous general-purpose AI agent. Within the code, references to a model dubbed Sierra were reportedly found. The references reportedly mentioned the Manus browser agent, which indicates that the company might be using the acquired startup's technology to integrate a browser agent. Its functionalities also remain unclear.

The tech giant is reportedly also planning to add support for OpenClaw (formerly Clawd and Moltbot). The references to the integration spotted by the publication in the code indicate that Meta AI will let users connect any model via the user's application programming interface (API) key, and the entire experience is being referenced to as an “OpenClaw agent”. However, just like the previous updates, it is unclear how this setup would work.

The abovementioned updates were based on references in code snippets of Meta AI, and it does not automatically indicate that these features will be shipped to users, or in the form that is currently seen in the code. Hence, readers should take the information with a pinch of salt until Meta officially announces the capabilities.