Meta Superintelligence Labs has announced the release of Muse Spark 1.1, a significant upgrade to its multimodal reasoning model designed for agentic tasks. The new model brings major improvements in tool and computer use, coding, and multimodal understanding, pushing the performance-efficiency frontier.
Muse Spark 1.1 is now available in “Thinking” mode on the Meta AI app and meta.ai, and developers can access it through the new Meta Model API, now in public preview. The model is built to handle complex agentic workflows, including planning, orchestration across external apps, and multi-agent delegation.
Key technical advancements include a 1 million token context window that actively manages memory, compacts critical steps, and retrieves information from earlier work. The model excels at computer-use tasks, adapting to changing interfaces and automating actions via scripting or direct interaction as needed.
In coding, Muse Spark 1.1 shows substantial gains on real-world tasks such as debugging, feature implementation, and large code migrations. It supports popular agentic coding setups with features like planning mode, subagent delegation, and context compaction. Internal evaluations on Meta Internal Coding Bench show significant improvements over the previous version.
Multimodal capabilities are also enhanced, enabling the model to inspect visual and audio data, generate detailed captions, and execute agentic workflows that combine perception and action. For example, it can create Facebook Marketplace listings from smartphone video.
Safety evaluations conducted under Meta’s Advanced AI Scaling Framework indicate that Muse Spark 1.1 operates within safe margins across frontier risk categories, with improved resistance to jailbreaks and prompt injection attacks. The full safety report is available online.
Early partners have praised the model as a complete agentic foundation, highlighting its long context handling, strong coding and reasoning, and OpenAI-compatible API. Meta plans to continue releasing more capable models in the future.