Meta Introduces a Big New AI Model for the Agentic Age

Meta has opened a public preview of Muse Spark 1.1, its largest and most powerful AI model yet, signaling a major push into the agentic era where AI systems act autonomously on behalf of users. The move places Meta in direct competition with OpenAI, Google, and Anthropic to dominate the next wave of artificial intelligence.

Meta unveiled the public preview of Muse Spark 1.1 on Wednesday, making the model available to developers worldwide. The company describes Muse Spark as its most advanced general-purpose AI foundation model, designed to power autonomous agents that can plan, reason, execute tasks, and adapt to new information without human oversight. Unlike earlier models that primarily generated text or images, Muse Spark is optimized for agentic workflows: it can browse the web, use APIs, write and execute code, and manage long-running sequences of actions.

The release comes as Meta accelerates its AI ambitions after a series of smaller open-source models like Llama 2, Llama 3, and Llama 3.1. While those were aimed at developers building chatbots and content tools, Muse Spark targets enterprise use cases such as automated customer service, supply chain optimization, and software development. The shift reflects a broader industry pivot from generative content to autonomous agents, with companies racing to build the operating system for the agentic age.

Muse Spark 1.1 is built on a transformer architecture with an estimated 1.2 trillion parameters, making it one of the largest publicly available AI models. It was trained on a curated dataset of over 20 trillion tokens, including web text, code repositories, scientific papers, and licensed content. Meta claims the model achieves state-of-the-art results on benchmarks for reasoning, tool use, and multi-step task completion. The preview allows developers to test the model through API endpoints and fine-tune it using Meta's custom hardware, the Meta Training and Inference Accelerator (MTIA) chips.

Meta CEO Mark Zuckerberg has framed Muse Spark as a foundational piece for the company's vision of AI-powered assistants that handle daily tasks across its apps. The model is designed to integrate with Meta's ecosystem—Facebook, Instagram, WhatsApp, and Ray-Ban Meta smart glasses—but also works independently for third-party developers. Analysts note that by offering a powerful agentic model openly, Meta hopes to attract a developer community that rivals OpenAI's ChatGPT platform and Google's Gemini suite.

The timing of the preview is strategic. The agentic AI market is expected to surpass $50 billion by 2027, according to Gartner. Companies like Microsoft, Amazon, and Salesforce are already embedding agentic capabilities into their products. Meta's late entry could be a challenge, but its open-access strategy and massive user base across its social platforms give it a unique distribution advantage. Privacy advocates, however, have raised concerns that an autonomous model deeply tied to Meta's advertising and data-collection infrastructure could lead to new forms of surveillance or manipulation.

Looking ahead, Meta plans to release a more refined version of Muse Spark later this year, along with a suite of safety tools to monitor and constrain agent behavior. The company also announced a partnership with Hugging Face to host community versions of the model, aiming for transparency and reproducibility. Developers can sign up for the public preview starting today. The success of Muse Spark will depend on whether Meta can convince developers and enterprises that agentic AI can be both powerful and trustworthy—a balance that has eluded even the most advanced models so far.