Mahalagang impormasyon
On October 12, Meta introduced the Muse Wonder and Muse Wonder Full AI models. These new additions to the Muse series prioritize practical agentic capabilities over visual demonstrations, such as 3D games, which Meta's Chief AI Officer Alexandr Wang noted were not a primary focus during development.
The models feature Active Memory, which allows the AI to identify and retrieve relevant information from past conversations, and Active Focus, a mechanism designed to persist through tasks until completion. These capabilities are supported by training on the Prometheus data center, utilizing a larger foundation model than the previous Spark iteration.
Performance benchmarks show mixed results. While Muse Wonder and Full outperformed Claude Opus 5.5 and GPT-6 Astra in MemoryBench and long-form coding tests like DeepSWE, they trailed competitors in specific metrics like ProgramBench and certain Artificial Analysis総合指標 scores. Meta's internal Personal Superintelligence Index places Wonder Full at 59.1, surpassing rivals, though the standard Wonder model scores 54.3.
Muse Wonder is currently available via the Meta Model API and Muse Code, with costs set at 4 dollars per million input tokens. Muse Wonder Full is accessible through Muse Code subscription plans starting at 50 dollars per month. CEO Mark Zuckerberg intends to expand access to these intelligence capabilities to personal AI agent users soon.
