Meta Superintelligence Labs announced the public release of Muse Spark 1.1, an updated multimodal reasoning model engineered to support autonomous, agentic workflows. The upgraded model features a one-million-token context window and includes significant improvements in coding performance and direct computer control capabilities.
The company said the new release is designed to operate natively within multi-agent networks. Muse Spark 1.1 can act both as a "primary agent" and as a "subagent," allowing the model to decompose large tasks and distribute work to parallel subagents. Meta describes this architecture as a way to reduce processing latency for complex, distributed workloads.
Muse Spark 1.1 is accessible to consumers in "Thinking" mode via the Meta AI app and on meta.ai, according to the announcement. In parallel with the consumer rollout, Meta published a public preview of its Meta Model API, which for the first time opens the Muse architecture to outside developers.
Meta presented Muse Spark 1.1 as a strategic step toward the enterprise AI market, emphasizing improvements in handling enterprise-scale codebases. Internal evaluations on Meta Internal Coding Bench reportedly show the model outpacing its predecessor and running competitively with unnamed industry alternatives. Meta said Muse Spark 1.1 can autonomously diagnose bugs, add new features into legacy enterprise code, and perform large-scale code migrations.
Beyond code generation and analysis, Muse Spark 1.1 introduces expanded "computer use" abilities. Meta stated the model can automate intricate workflows spanning multiple desktop applications by either authoring automation scripts or interacting directly with graphical user interfaces when text-based scripting is not feasible.
A trio of early API partners - Replit, Cline, and Box - were highlighted at the announcement, signaling immediate adoption interest from developer tooling providers. Amjad Masad, CEO of Replit, said that "The combination of a million-token context window, deep multimodal support, and advanced coding capabilities fundamentally changes how we think about scale."
Saoud Rizwan, CEO of Cline, commented on the model’s tool-use flexibility and competitive pricing for high-volume developer workloads. Yashodha Bhavnani, VP of AI Products at Box, noted that Spark 1.1 performed head-to-head with top-tier frontier models on Box’s internal enterprise evaluation sets.
Meta also addressed safety concerns tied to autonomous AI systems. The company reported that Muse Spark 1.1 underwent safety auditing under its Advanced AI Scaling Framework. Evaluations focused on catastrophic thresholds including chemical and biological risks, cybersecurity vulnerabilities, and "loss of control" scenarios. Meta stated the model operates within its established safety margins and said it will publish a comprehensive evaluation report in the near term.
The launch of Muse Spark 1.1 coincides with the general release of Muse Image, Meta’s standalone image-generation model. Meta framed the simultaneous rollouts as part of a broader update to its Muse family of models.
Implications for markets and sectors
- Enterprise software and developer tooling providers may see increased demand for integration with Muse Spark 1.1 following the Meta Model API public preview.
- Companies that rely on large-scale code maintenance, legacy system upgrades, and automation of desktop workflows could be directly impacted by Muse Spark 1.1’s coding and computer-control features.
- The rollout may affect cloud and infrastructure usage patterns where multi-agent, low-latency processing and large context windows are required.
Technical highlights
- One-million-token context window to support large context and enterprise-scale tasks.
- Native multi-agent operation with roles as primary agent and subagent to enable parallel task delegation and lower latency.
- Enhanced coding capabilities validated on Meta Internal Coding Bench and claimed to be competitive with leading frontier models.
- New computer-control features that can author automation scripts or interact with application user interfaces directly.