News analysis · Published
Meta Muse Wonder: Separate Memory From Action Authority
By the ELYMENT AI editorial team · Free to read
Meta released Muse Wonder and Muse Wonder Full on 12 October 2026 for long-running agent, coding and computer-use tasks. Muse Wonder is available through the Meta Model API and Muse Code, while the Full model is limited to paid Muse Code plans. The useful business test is not whether the model remembers more. It is whether memory can inform work without silently expanding the agent’s authority to disclose data, change records or complete irreversible actions. [1][2]
What Meta Muse Wonder changes
Meta describes Active Memory as carrying important context across conversations and retrieving it when relevant. Active Focus is intended to keep an agent working through blockers, trying alternatives and checking in before important actions. The models also target long-running software work and computer use across browsers, desktops and mobile apps. [1][2]
Muse Wonder accepts text, image, video and audio input, returns text and supports a one-million-token context window. Meta lists the API model at US$4 per million input tokens, US$0.20 per million cached input tokens and US$20 per million output tokens. Muse Wonder Full is not offered through the API; it is included in Muse Code’s US$50 and US$200 monthly plans. [2][3]
Why memory and authority need separate controls
Persistent memory can reduce repeated briefing, but it also increases the chance that old, sensitive or superseded information influences a new task. A useful memory is not automatically an authorised fact, and a remembered preference is not permission to send, buy, edit or disclose.
Meta’s model card says performance can vary with prompts, tools, serving configuration and the agent harness. It also warns that the models can be inaccurate and may make errors in novel multi-step scenarios. Those limits mean a benchmark result cannot substitute for testing the full business system, including connectors, permissions, audit logs and recovery. [3]
Use a four-gate business trial
Start with a contained workflow where the model can read approved material but cannot take consequential action. Score four gates before expanding access:
- Memory: confirm what is retained, when it is retrieved, how updates supersede old facts and how data is deleted.
- Authority: map every tool action to an explicit permission, with fresh confirmation for external messages, purchases and destructive changes.
- Evidence: require source links, action previews and a complete record of prompts, tool calls, approvals and outcomes.
- Recovery: test cancellation, credential revocation, rollback and incident review before the agent reaches production data.
What business leaders should do next
Choose one reversible process, a named owner and a short test set drawn from real exceptions rather than a polished demo. Compare Muse Wonder with the current system on completion quality, human review time, unauthorised-action rate and total cost. Keep memory scope narrow until the team can prove that corrections propagate and permissions remain bounded.
Meta’s release adds a credible new option for agentic work, not a reason to relax operating discipline. ELYMENT AI helps teams turn model capability into governed workflows with explicit context, approvals and recovery paths.
Sources
- Meta AI Research: Muse Wonder (2026-10-12) - Meta’s research announcement covering Active Memory, Active Focus, agentic capabilities and availability.
- Meta developer page: Muse Wonder (2026-10-12) - Official model access, context, pricing, plan and benchmark information.
- Muse Wonder Model Card (2026-10-12) - Official model specifications, evaluation methodology, limitations and responsible-use guidance.
- The Verge: Meta has some Wonder Full new AI models (2026-10-12) - Independent reporting on availability, competitive claims and the planned Muse integration.
Continue learning
Related analysis
- Meta Muse Zero-Day: Secure the Agent Client, Not Just the Cloud
Meta patched a Muse Mac client flaw that let local code redirect agent traffic. Businesses should secure endpoints, tokens, permissions and audit paths.
- Meta Muse Launch: What Employers Should Control Before Personal Agents Reach Work
Meta Muse can connect to email, calendars and payments. Learn how employers should govern identity, data, approvals and revocation before work use.
- Meta Muse Spark 1.3: What Agentic AI Buyers Should Test
Meta released Muse Spark 1.3 for coding and agentic workflows. Buyers should test task success, tool discipline, cost, data terms and recovery.
Frequently asked questions
What is Meta Muse Wonder?
Muse Wonder is Meta’s frontier model for coding, long-running agents and computer use. It is available through the Meta Model API and Muse Code.
How is Muse Wonder Full different?
Meta positions Muse Wonder Full as the more capable variant. It is currently available through paid Muse Code plans rather than the Meta Model API.
How should a business evaluate Muse Wonder?
Test the complete workflow across memory, authority, evidence and recovery. Keep consequential actions behind human confirmation until real operating evidence supports wider access.