Muse Wonder Full (Max)

Muse Wonder Full (Max) is a frontier large language and multimodal model developed by Meta Superintelligence Labs, released on October 12, 2026. Built as the flagship variant of the Muse Wonder series, it is engineered for advanced personal and professional agentic tasks, achieving competitive performance on reasoning, coding, and multi-step workflows. Trained at Meta's Prometheus data center, the model is proprietary and accessed via platform subscriptions rather than open weights.

A defining characteristic of the Muse Wonder Full model is its integration of Active Memory and Active Focus. Active Memory enables the model to retain context and critical details across extended conversations spanning days or weeks, while Active Focus allows the system to persist through execution challenges, experiment with alternative strategies, and recover from errors during long-horizon tasks.

Capabilities and Context Window

The model supports a context window of approximately one million tokens and natively processes text, images, audio, and video inputs while generating text outputs. Muse Wonder Full demonstrates strong competency in expert domains such as legal reasoning, financial analysis, and complex programming, landing near the top of multiple independent evaluation leaderboards for agent-driven execution.