Meta has released Muse Spark 1.1, a new multimodal reasoning model from Meta Superintelligence Labs designed for agentic tasks, coding, computer use and long running workflows.

The model is available through Meta’s new Meta Model API, now in public preview, and is also live in Thinking mode inside the Meta AI app and on meta.ai.

Mark Zuckerberg said on X that Muse Spark 1.1 is “a strong agentic and coding model at a very low price,” adding that it is built for tool use, computer use and parallel subagent execution.

(1) Today we're releasing Muse Spark 1.1 — a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.

— Mark Zuckerberg (@finkd) July 9, 2026

Muse Spark 1.1 is positioned as a major upgrade from the original Muse Spark. Meta says the model can plan across external apps and services, generalize to new native tools and MCP servers, and coordinate multiagent systems to reduce latency on complex tasks.

The company also said the model supports a 1 million token context window, allowing it to retain earlier actions, retrieve information from long workflows and compress context while preserving critical steps.

Computer use is one of the main areas of focus. Meta said Muse Spark 1.1 can work across desktop, mobile and browser interfaces, deciding when to automate with scripts and when to interact directly with an interface.

The company cited examples such as organizing a dinner party and adjusting an order when new information appears during the task.

Coding performance also received a major upgrade. Meta said Muse Spark 1.1 can diagnose bugs, implement features in large codebases and support large code migrations. The company said developers and researchers inside Meta are already using the model for coding, model development and evaluation workflows.

The launch also pushes Meta further into multimodal agents. Muse Spark 1.1 can process images, video, audio and PDFs, and use that context while operating tools or computers.

Meta pointed to use cases such as generating code from visuals, creating detailed image and video captions, and using smartphone footage to build a Facebook Marketplace listing on a user’s behalf.

Meta said the model passed its safety evaluations under the company’s Advanced AI Scaling Framework, including reviews across chemical and biological risk, cybersecurity and loss of control. The company also said Muse Spark 1.1 showed stronger resistance to jailbreaks, prompt injection and other adversarial attacks.

The release gives developers direct access to Muse Spark for the first time through an OpenAI compatible API. Replit CEO Amjad Masad described the model as a “complete agentic foundation,” while Cline CEO Saoud Rizwan said the combination of strong tool use and low cost made it viable for large scale coding workloads.

The announcement comes alongside Meta’s launch of Muse Image, extending the company’s push into agentic and multimodal AI systems. Zuckerberg said Meta’s focus is on delivering strong agentic and multimodal models at very low cost, with more models to come.