Muse Wonder

// Overview

Muse Wonder is a model series from Meta Superintelligence Labs, released on October 12, 2026 in two settings: Muse Wonder and Muse Wonder Full. Meta describes them as its most capable models, built on the groundwork of Muse Spark and aimed at personal and professional agentic tasks such as handling an insurance claim, understanding a lease or building a working app, with a focus on law, finance, health and frontier coding.

Two trained behaviours define the series. Active Memory carries context about the user across conversations and retrieves the details that matter for the current task. Active Focus keeps the model working toward a goal: it tries alternative approaches, works through blockers, adapts to new information, asks for clarification when needed and checks in before important actions. Meta says the models were pretrained and trained with reinforcement learning at its Prometheus data center, then post-trained on diverse, long-running tasks.

Muse Wonder (model id muse-wonder-1.0) is served through the Meta Model API and Muse Code with a 1M-token context window, at $4.00 per million input tokens, $0.20 per million cached input tokens and $20.00 per million output tokens. Muse Wonder Full (muse-wonder-full-1.0) is available only inside Muse Code on a paid subscription, starting with the $50-a-month Power Usage plan. Meta releases these models under its Meta Superintelligence Scaling Framework.

// Pricing

muse-wonder-1.0 on the Meta Model API. Muse Wonder Full is not offered through the API; it is included in the Muse Code Power Usage ($50/month) plan.

Pricing source ↗

// Strengths

• Active Memory: carries user context across conversations and retrieves what is relevant to the current task
• Active Focus: keeps pursuing a goal through errors and blockers, asks for clarification and checks in before important actions
• Long-running agents: plans multi-step tasks, delegates to sub-agents, calls tools in parallel and recovers from errors
• Computer use across browsers, desktop and mobile apps
• 1M-token context window on both Muse Wonder and Muse Wonder Full

// Best for

• Reach for it for long agent runs where the model must hold user context and keep working through failures without hand-holding.
• Reach for it in Muse Code for multi-file work across large repositories, or Muse Wonder Full on a Power Usage plan for the heaviest jobs.
• Reach for Muse Spark instead when latency and cost per token matter more than maximum capability.

// How to access