New Delhi: Meta on Monday launched Muse Wonder and Muse Wonder Full, two premium artificial intelligence (AI) models designed for complex personal and professional tasks, including software development, financial and legal analysis, and healthcare-related information.
Developed by Meta Superintelligence Labs, the models build on the Muse Spark model launched earlier this year.
The new models introduce two key capabilities: 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 also seek clarification and check with users before taking important actions.
Meta said Muse Wonder is available through its Model API and Muse Code, while Muse Wonder Full is restricted to Muse Code subscribers.
The company plans to integrate Muse Wonder into its Muse AI experiences, initially rolling it out to subscribers.
Reported API pricing is $4 per million input tokens and $20 per million output tokens, while paid Muse Code plans range from $50 to $200 per month.
In benchmark results shared by Meta, Muse Wonder Full scored 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, Wonder Full scored 79.0, narrowly ahead of Wonder at 78.8.
Meta said the models offer advanced coding performance and stronger safeguards against chemical, biological, radiological, nuclear and cyber misuse, while retaining capabilities for defensive cybersecurity.
The launch comes amid growing competition in advanced AI and questions about privacy and security surrounding AI systems with persistent memory. Meta said the models were trained at its Prometheus data centre in central Ohio.
Clarify the launch dateAdd Meta’s stated evaluationsTighten the pricing attribution
