What changed
Meta released Muse Wonder through the Meta Model API and Muse Code, with a one-million-token context window and listed API pricing of $4 per million input tokens and $20 per million output tokens. The larger Muse Wonder Full is available only to Muse Code subscribers. Meta positions both models for coding and sustained personal and professional agent tasks with Active Memory and Active Focus.
Why it matters
This is a new hosted frontier-model choice for long-running agents rather than a research preview. Its cross-conversation memory, browser and application work make model quality, retained personal context, tool authority and cost part of one deployment decision.
What to do
Canary Muse Wonder on one reversible, low-privilege agent workflow. Measure task completion, latency and token cost; inspect every tool call; test clarification, denied actions and memory deletion; and compare against the currently deployed model before allowing access to messages, calendars, documents or consequential writes.
Caveat
Availability and benchmark results are Meta-reported, and INTELLIAI did not independently reproduce them. Meta says a fuller safety report with additional evaluations will follow. Muse Wonder Full has no general API availability, while persistent memory and browser or app access increase privacy and authorization risk.