Meta on October 12, 2026 introduced the Muse Wonder series—Muse Wonder and Muse Wonder Full—describing them as its most capable models for personal and professional agentic work and as the next step on its scaling ladder toward personal superintelligence. In its primary research announcement, Meta said Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full is available for Muse Code subscribers. Meta also said Muse Wonder will soon power experiences in Muse, rolling out first to subscribers, and will support artifact generation and higher-intelligence processes.
The announcement highlights two agent-oriented capabilities: Active Memory, which Meta defines as carrying important context about a user across conversations and retrieving details relevant to the current task, and Active Focus, which it describes as sustained work toward a challenging goal—trying alternatives, working through blockers, adapting to new information, asking for clarification when needed, and checking in before important actions. Meta frames domain expertise around law, finance, and health evaluations, including Harvey’s Legal Agent Benchmark (LAB), and points to FocusBench and SWE-Marathon for sustained execution on long-running software projects. Frontier coding is presented as the place those capabilities combine, with both models trained on diverse long-running tasks.
Meta said the models were trained at its gigawatt-scale Prometheus data center and build on the scaling approach it outlined with Muse Spark. A companion developer post priced Muse Wonder on Muse Code and the Meta Model API at $4 input and $20 output per million tokens (USD), and said Muse Wonder Full is available exclusively by subscription in Muse Code on the Power Usage and new Maximum Usage plans. Artificial Analysis independently reported Intelligence Index v4.3 scores at max reasoning effort of 52 for Muse Wonder and 56 for Muse Wonder Full, versus 48 for Muse Spark 1.3 (max), and noted Muse Wonder Full is not priced per token. Artificial Analysis also listed 1M-token context windows, multimodal inputs (text, image, video, and audio), and text output.
On safety, Meta said the Muse Wonder family was trained for robustness to adversarial instructions and careful handling of private information, with what it calls its strongest safeguards to date against CBRN and cyber misuse while preserving defensive cyber capability under its Meta Superintelligence Scaling Framework. Meta said a fuller safety evaluation report will be available soon. Euronews and Artificial Analysis independently covered the October 12 launch, aligning on the two-model release, agentic positioning, and Artificial Analysis benchmark framing.
DigiEditorial verified the launch against Meta’s October 12 Muse Wonder research post as the primary source, with corroboration from Meta’s developer blog on Muse Wonder API and Muse Code pricing and from Artificial Analysis’s October 12 evaluation write-up. Capability claims, pricing, availability, and safety statements are company disclosures or independent reporting of published benchmark scores. DigiEditorial has not independently tested Muse Wonder or Muse Wonder Full.