What happened
News Mobile reports that Meta launched two premium AI models, Muse Wonder and Muse Wonder Full, developed by Meta Superintelligence Labs. The models reportedly add Active Memory for retaining relevant information across conversations and Active Focus for exploring alternatives, recovering from obstacles and adapting to new information. Muse Wonder is reportedly available through Meta’s Model API and Muse Code, while Muse Wonder Full is limited to Muse Code subscribers. The report identifies the launch as occurring on Monday, but does not provide an exact calendar date.
News Mobile reports that Meta launched Muse Wonder and Muse Wonder Full on Monday. The supplied report does not state the exact calendar date, although the surrounding context places the report within the current news window. Meta reportedly describes the models as premium systems for software development, financial and legal analysis, healthcare-related information and other complex personal or professional tasks.
The report says the models introduce Active Memory, which retains and retrieves relevant information across conversations, and Active Focus, which explores alternative approaches, recovers from technical obstacles and adapts to new information. They can reportedly ask for clarification and check with users before taking important actions.
According to News Mobile, Muse Wonder is available through Meta’s Model API and Muse Code, while Muse Wonder Full is restricted to Muse Code subscribers. The report gives API pricing of $4 per million input tokens and $20 per million output tokens, and says paid Muse Code plans range from $50 to $200 per month. These prices are attributed to reporting and are not independently confirmed.
Meta reportedly shared evaluation results showing Muse Wonder Full at 13.1 on Harvey LAB-AA, compared with 10.0 for Muse Wonder and 8.6 for GPT-6 Astra. On the DeepSWE 1.1 long-horizon coding evaluation, the report says Wonder Full scored 79.0, narrowly ahead of Wonder at 78.8. The report does not provide methodology, test conditions or independent replication.
Source details: newsmobile.in ↗
Why it matters
If the reported capabilities and access terms are accurate, the launch would expand Meta’s effort to compete in advanced AI systems designed for longer, more complex personal and professional workflows. Persistent memory and the ability to pause for clarification could make these systems more useful for coding, analysis and other multistep work, while also increasing privacy and control risks. The report’s performance figures come from evaluations shared by Meta and are not independently verified.
The reported release is consequential because it combines stronger long-horizon task behavior with persistent cross-conversation memory. That combination could improve continuity in coding, research and professional workflows, but it also raises the stakes of incorrect recall, stale information and inappropriate action.
The models’ reported ability to explore alternatives and recover from obstacles suggests a focus on extended agent-like work rather than one-turn responses. However, scores alone do not establish reliability in real-world financial, legal, healthcare or cybersecurity use.
Persistent memory creates practical questions about consent, retention, deletion, access controls and the separation of information between users or projects. The supplied report mentions privacy and security concerns but does not provide Meta’s detailed policies or technical safeguards.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
What to watch next
The exact launch date, availability by country and user tier, official API documentation, and whether the reported pricing is current remain unclear. Users and developers should also look for independent testing of the models, details about data retention and memory controls, and evidence that safeguards work without materially reducing defensive cybersecurity capabilities.
Confirm the exact launch date and whether Muse Wonder is generally available or limited by geography, account type, quota or invitation. The report does not establish broad availability.
Verify the reported API and Muse Code prices against Meta’s official documentation. No source material supplied here confirms billing units, included usage, rate limits or whether Muse Wonder Full has separate charges.
Independent evaluations should test coding reliability, error recovery, memory accuracy, clarification behavior and safety performance. Meta’s reported results are company-provided and cannot by themselves establish comparative performance.
Further documentation is needed on how persistent memory is stored, controlled, audited and deleted, especially for sensitive professional or healthcare-related information.