Meta today announced Muse Spark 1.1, a new version of its AI model focused on agentic AI, software development, and advanced computer control. At the same time, the company is launching the new Meta Model API, giving developers their first chance to build apps on Muse Spark. CEO Mark Zuckerberg revealed the news Thursday on X.
According to Zuckerberg, Muse Spark 1.1 is built for complex, multi-step tasks that AI can execute autonomously. The model features a one-million-token context window, can split work across parallel sub-agents, and is trained to control computer interfaces on desktop, mobile, and in web browsers on its own.
Agentic AI and coding take center stage
Meta is positioning Muse Spark 1.1 squarely as a model for the next wave of AI assistants. Where traditional chatbots mostly generate answers, agentic AI is designed to run entire workflows. That means a model can use external tools, make decisions, and chain actions without constant user input.
Zuckerberg says Muse Spark 1.1 excels at:
- agentic tasks;
- software development and programming;
- using external tools;
- controlling computers and applications;
- sustained workflows via a 1 million-token context window.
Support for parallel sub-agents should make the model more efficient on large jobs, letting the AI tackle parts of a task simultaneously and then merge the results.
New Meta Model API opens the door for developers
Alongside the model, Meta is introducing the Meta Model API. This interface lets developers integrate Muse Spark into their own apps, AI agents, and enterprise software.
Zuckerberg says Meta aims to stand out by offering powerful multimodal and agentic models at a relatively low price, moving more directly into the commercial API market dominated by OpenAI, Anthropic, and Google.
According to Reuters, the API is currently in a developer preview. Meta is offering free starter credits before a paid plan kicks in.
A strategic play in the AI race
Muse Spark 1.1 fits into Meta’s broader AI strategy. Earlier this year, the company launched Muse Spark as the first model from Meta Superintelligence Labs, the new AI division led by Alexandr Wang. Since then, Meta has also released Muse Image and a preview of Muse Video under the same model family.
With the new API, Meta is leaning further into commercial AI services. After years of being known for its open Llama models, it’s now adding a paid API that gives developers direct access to its latest systems. Reuters frames this as a head-to-head push against providers like OpenAI and Anthropic.
Limited access at launch
While Zuckerberg said on X that Muse Spark 1.1 is available via the Meta Model API, several developers quickly noted that access still appears limited for now.
Meta has not offered detailed clarification during the announcement. Reuters reports the rollout is starting as a preview before broader availability.
Why it matters
Muse Spark 1.1 underscores how competition among major AI players is shifting from simple chatbots to autonomous AI agents capable of handling complex tasks. Price is also becoming a key differentiator. Meta is betting on relatively low-cost API access combined with robust agentic capabilities.
For developers, the new Model API means they can now build directly on Meta’s latest AI tech alongside existing platforms from OpenAI, Anthropic, and Google.