Meta Platforms’ AI division, Meta Superintelligence Labs, announced the release of Muse Spark 1.1 on Thursday. The new multimodal reasoning model is designed for autonomous agentic workflows and comes with a one-million-token context window.
The update also brings major improvements in coding, multi-step reasoning, and direct computer control.
Meta Targets Enterprise AI With Muse Spark 1.1
The launch marks another major step in Meta’s push into the enterprise AI market. Alongside the consumer release, Meta also introduced a public preview of its Meta Model API.
This gives outside developers access to the Muse model architecture for the first time.
Muse Spark 1.1 is already available in “Thinking” mode through the Meta AI app and on meta.ai.
Built for Multi-Agent AI Workflows
One of the key features of Muse Spark 1.1 is its ability to work inside multi-agent systems.
According to Meta, the model can act both as a main agent and as a subagent. This allows it to divide large tasks into smaller parts and assign them to parallel subagents.
As a result, the model can process complex workflows faster and reduce delays in large-scale AI tasks.
Stronger Coding and Enterprise Software Support
Meta is positioning Muse Spark 1.1 as a serious rival to leading AI coding models.
The company says the model performs better on large enterprise codebases and can handle complex software development tasks. These include finding bugs, adding new features to older systems, and managing large code migrations.
Meta’s internal coding tests showed that Muse Spark 1.1 outperformed its earlier version and performed competitively against other major industry models.
New Computer Use Capabilities
Muse Spark 1.1 is not limited to writing code. The model also includes new computer use features that allow it to automate tasks across desktop applications.
It can create automation scripts when possible. When scripting is not enough, it can also interact directly with user interfaces.
This makes the model more useful for practical workflows that involve several apps, tools, or enterprise systems.
Early Partners Signal Developer Interest
Meta also highlighted several early API partners, including Replit, Cline, and Box.
These partnerships suggest that Muse Spark 1.1 is already gaining attention in the developer tools market.
Replit CEO Amjad Masad said the model’s large context window, multimodal support, and coding ability could change how developers think about scale. Cline CEO Saoud Rizwan highlighted the model’s flexible tool use and competitive pricing for heavy developer workloads.
Box also reported strong results, with VP of AI Products Yashodha Bhavnani saying Spark 1.1 performed well against top frontier models in the company’s internal enterprise tests.
Meta Highlights AI Safety Testing
Meta also addressed safety concerns around autonomous AI systems.
The company said Muse Spark 1.1 was tested under its Advanced AI Scaling Framework. These evaluations focused on serious risk areas, including cybersecurity, chemical and biological threats, and possible loss-of-control scenarios.
Meta said the model remained within its safety limits. A full evaluation report is expected to be released soon.
Muse Image Launches Alongside Spark 1.1
The release of Muse Spark 1.1 comes at the same time as the general launch of Muse Image, Meta’s latest standalone image-generation model.
Together, the two releases show Meta’s broader strategy to compete across multimodal AI, enterprise automation, coding tools, and generative media.
