Meta launched Muse Spark 1.1 on July 9, 2026, positioning the model as its most capable system to date and a direct challenge to frontier offerings from OpenAI and Anthropic. The release follows the original Muse Spark introduction earlier in 2026 and represents a significant upgrade across reasoning, multimodal understanding, and agentic performance. Meta framed the launch as a strategic step toward delivering what it calls “personal superintelligence” to end users and enterprises.
Meta’s Muse Spark 1.1 challenges OpenAI and Anthropic as the company pursues its vision of personal superintelligence.
Muse Spark 1.1 accepts text, images, video, audio, and PDF inputs, producing text-only outputs. The model was built specifically for agentic tasks, including planning, task decomposition, and orchestration across external applications and services. Meta designed the architecture to support complex multistep workflows and digital process management in enterprise environments, with particular emphasis on perception-heavy tasks that require coordinating multiple tools simultaneously.
A central feature of Muse Spark 1.1 is its support for multi-agent automation workflows, where multiple specialized agents execute subtasks in parallel under a higher-level controller. This parallel subagent orchestration capability marks what Meta describes as a major step change from the first-generation model. The system also supports computer use features, allowing it to operate user interfaces and tools as part of agentic task flows.
Meta additionally reported zero-shot generalization to its tool and computer use protocol, reducing fine-tuning requirements when deploying the model against new tools.
On coding performance, Meta reported that internal benchmark scores on its “atomic suite” increased from 48.1% to 67.0% pass@1 between Muse Spark and Muse Spark 1.1. The model supports iterative debugging and self-correction, including the ability to repair its own generated code within a session. Meta positioned Muse Spark 1.1 as its strongest model for real-world software maintenance, bug fixing, and end-to-end software delivery pipelines, competing directly with coding-focused models from OpenAI and Anthropic in enterprise settings.
The agentic improvements were partly shaped by feedback from early developer partners, and the model’s tooling gains complement broader enhancements to multimodal understanding that distinguish it from its predecessor. Meta indicated that Muse Spark 1.1 is particularly well suited for personal agentic tasks requiring coordination across multiple apps and services, a use case the company sees as central to its broader AI product direction.
Meta Superintelligence Labs developed Muse Spark 1.1 as part of a wider effort to close competitive gaps in advanced AI capabilities and commercial AI business models. The launch signals Meta’s intent to compete not only on open model releases but also in the enterprise and developer markets where OpenAI and Anthropic have established stronger footholds. The model is capable of managing a context window of 1 million tokens, enabling it to retain and retrieve information across extended, complex sessions.
Muse Spark 1.1 represents the company’s clearest attempt yet to convert its research investments into a commercially competitive frontier model.
