Competition is shifting toward useful agents. Meta positions Wonder and Wonder Full as its most capable models, built for personal and professional tasks such as navigating insurance claims and interpreting documents, rather than just answering questions. 511 Strategically, the relevant test is whether that intelligence reliably completes work.

Premium access matters more than unrestricted distribution. The retrieved reporting says Wonder will initially power Muse features for subscribers, while Wonder Full is restricted to paid coding subscribers; separate reporting says neither model has publicly downloadable weights and only the standard model is available through the API. 51015 That points toward controlled commercialization, although the available excerpts do not fully verify the precise Meta Model API and Muse Code access arrangements.

The performance story remains a company claim. Reporting describes Meta’s models as approaching Anthropic’s strongest systems on reasoning, writing and coding, while another retrieved report attributes selected legal, financial and coding benchmark leads over Claude Opus 5.5 and GPT-6 Astra to the “Watermelon” model. 612 These descriptions are compatible with task-specific strengths, not proof of across-the-board superiority; independent tests must check comparable settings, costs and real-world task completion.

Memory and sustained effort are central to the agent proposition. Retrieved reporting identifies Active Memory and Active Focus as features supporting long-running, multistep workflows. 10 The strategic implication is that Meta wants assistants that retain context and persist through tasks—but those same capabilities make consent, data retention and authorization more consequential.

Muse is becoming an ecosystem, not just an app. Zuckerberg called it the “centrepiece” of his AI vision, and the September 23 Charm announcement added a dedicated physical access point that bypasses smartphone AI apps. 12 Meta also expanded Muse toward small businesses through Shopify, QuickBooks, Stripe and Canva integrations, explicitly seeking revenue beyond advertising. 3

Early adoption supports optimism, not a revenue conclusion. Citi projected more than $27 billion in Muse revenue by 2030, while reporting of its analysis cited 6.6 million downloads and 1.8 million daily users. 47 That is an analyst forecast, not realized revenue; the decisive questions are paid conversion, retention, agent reliability and the cost of serving each user.

The privacy concern is substantive. TIME’s retrieved investigation says Muse maps users and their contacts—including nonusers—and shares anonymized insights across its system, based on internal instructions. 6 That raises questions about third-party consent and whether de-identification adequately protects sensitive relationship and behavioral information; the available excerpt does not independently establish the claimed hourly update frequency.

Important details remain unverified. Insufficient evidence in the supplied excerpts to substantiate the exact 14-month hiring-and-restructuring account, Monday’s share-price move or the rally toward a $2 trillion valuation, Prometheus training details, the Scaling Framework’s specific safeguards, planned safety disclosures, testing misbehavior, Zuckerberg’s safety-policy position, or Wang’s precise open-source commitment.

The same limitation applies to several Muse product details. Insufficient evidence here to verify the complete free/$20/$100 tier structure, WhatsApp availability, dedicated virtual machines, the full email-to-payment capability list, App Store ranking, weekly-user figure, or reports that Zuckerberg pushed launch despite internal privacy objections.

Competition is shifting toward useful agents. Meta positions Wonder and Wonder Full as its most capable models, built for personal and professional tasks such as navigating insurance claims and interpreting documents, rather than just answering questions. 511 Strategically, the relevant test is whether that intelligence reliably completes work.

Premium access matters more than unrestricted distribution. The retrieved reporting says Wonder will initially power Muse features for subscribers, while Wonder Full is restricted to paid coding subscribers; separate reporting says neither model has publicly downloadable weights and only the standard model is available through the API. 51015 That points toward controlled commercialization, although the available excerpts do not fully verify the precise Meta Model API and Muse Code access arrangements.

The performance story remains a company claim. Reporting describes Meta’s models as approaching Anthropic’s strongest systems on reasoning, writing and coding, while another retrieved report attributes selected legal, financial and coding benchmark leads over Claude Opus 5.5 and GPT-6 Astra to the “Watermelon” model. 612 These descriptions are compatible with task-specific strengths, not proof of across-the-board superiority; independent tests must check comparable settings, costs and real-world task completion.

Memory and sustained effort are central to the agent proposition. Retrieved reporting identifies Active Memory and Active Focus as features supporting long-running, multistep workflows. 10 The strategic implication is that Meta wants assistants that retain context and persist through tasks—but those same capabilities make consent, data retention and authorization more consequential.

Muse is becoming an ecosystem, not just an app. Zuckerberg called it the “centrepiece” of his AI vision, and the September 23 Charm announcement added a dedicated physical access point that bypasses smartphone AI apps. 12 Meta also expanded Muse toward small businesses through Shopify, QuickBooks, Stripe and Canva integrations, explicitly seeking revenue beyond advertising. 3

Early adoption supports optimism, not a revenue conclusion. Citi projected more than $27 billion in Muse revenue by 2030, while reporting of its analysis cited 6.6 million downloads and 1.8 million daily users. 47 That is an analyst forecast, not realized revenue; the decisive questions are paid conversion, retention, agent reliability and the cost of serving each user.

The privacy concern is substantive. TIME’s retrieved investigation says Muse maps users and their contacts—including nonusers—and shares anonymized insights across its system, based on internal instructions. 6 That raises questions about third-party consent and whether de-identification adequately protects sensitive relationship and behavioral information; the available excerpt does not independently establish the claimed hourly update frequency.

Important details remain unverified. Insufficient evidence in the supplied excerpts to substantiate the exact 14-month hiring-and-restructuring account, Monday’s share-price move or the rally toward a $2 trillion valuation, Prometheus training details, the Scaling Framework’s specific safeguards, planned safety disclosures, testing misbehavior, Zuckerberg’s safety-policy position, or Wang’s precise open-source commitment.

The same limitation applies to several Muse product details. Insufficient evidence here to verify the complete free/$20/$100 tier structure, WhatsApp availability, dedicated virtual machines, the full email-to-payment capability list, App Store ranking, weekly-user figure, or reports that Zuckerberg pushed launch despite internal privacy objections.