Meta Superintelligence Labs has released Muse Spark 1.1, an updated multimodal reasoning model designed for agentic tasks, with reported improvements in tool use, computer use, coding, and multimodal understanding.

The model is available in "Thinking" mode on the Meta AI app and on meta.ai. Alongside the release, Meta launched a public preview of the Meta Model API, giving developers access to Muse Spark 1.1 for the first time.

Muse Spark 1.1 supports a context window of one million tokens and is built to orchestrate multi-agent systems, allowing it to function as both a primary agent and a subagent. The model can delegate tasks across parallel subagents to reduce end-to-end latency.

In coding, the model is said to handle large, complex codebases, including bug diagnosis, feature implementation in enterprise systems, and large-scale code migrations. Meta said internal evaluations on its Meta Internal Coding Bench show Muse Spark 1.1 outperforms its predecessor and is competitive with unspecified alternatives.

For computer use, the model can automate workflows across multiple applications, write scripts when automation is faster, and interact directly with interfaces when that approach is simpler.

Meta said it conducted safety evaluations under its Advanced AI Scaling Framework across categories including chemical and biological risks, cybersecurity, and loss of control, and that the model operates within defined safe margins. A full evaluation report is described as forthcoming.

Early API partners include Replit, Cline, and Box. Amjad Masad, CEO of Replit, cited the model's million-token context, multimodal support, and coding capabilities. Saoud Rizwan, CEO of Cline, noted its tool use and pricing for coding workloads at scale. Yashodha Bhavnani, VP of AI Products at Box, said the model performed competitively against frontier models on Box's enterprise evaluation set.

The release coincides with Meta's launch of Muse Image, according to the company statement.