Meta releases Muse Wonder and Muse Wonder Full AI models

Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence, Meta announced on Oct. 12, 2026, as the company released Muse Wonder and Muse Wonder Full through its Superintelligence Lab.

Meta released its first frontier artificial intelligence models on Oct. 12, 2026, introducing Muse Wonder and Muse Wonder Full to compete directly with offerings from OpenAI and Anthropic, according to reporting by TechCrunch and Economic Times. The release follows a 14-month overhaul of Meta’s artificial intelligence efforts led by Scale AI founder Alexandr Wang, who restructured the company’s research teams in August 2025.

Meta Superintelligence Lab Launches Wonder Models

The newly unveiled Wonder models are larger and more capable than Meta’s previous lightweight Muse Spark model, achieving higher benchmark scores on complex tasks such as coding. According to testing presented in Meta’s model card, Muse Wonder and Muse Wonder Full deliver benchmark results comparable to Anthropic’s Fable 5.1 and OpenAI’s GPT-6 Astra. The models were trained at Meta’s Prometheus data center in central Ohio, which became operational earlier in 2026.

Economic Times reported that Meta shares rose about 4% in afternoon trading following the release. The launch follows CEO Mark Zuckerberg pledging billions of dollars and aggressively poaching researchers over the past year to catch up in Silicon Valley’s intensifying artificial intelligence race.

Pricing and Distribution Across Meta APIs

Access to the new models varies by tier and service channel. Accessed through the Meta Model API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens, positioning it at a price point roughly equivalent to GPT-6 Sol or Opus 5.5.

Economic Times and TechCrunch confirmed that distribution differs between the two variants. Muse Wonder will soon power Meta’s Muse personal artificial intelligence agent for subscribers, while the more capable Muse Wonder Full model is restricted exclusively to subscribers of the Muse Code coding tool and will not be served through the standard API.

Active Memory Benchmarks and Recall Tests

Meta placed particular emphasis on the new models’ ability to retain and synthesize user information over extended periods, a property the company terms active memory. Research scientist Kavya Kopparapu detailed the memory engineering in a launch video, stating We put a tremendous amount of effort into making sure your model can remember what’s important to you over days, weeks, months, and beyond.

To measure this capability, Meta highlighted the MemoryBench benchmark, which tests continual learning and memory construction frameworks. Meta’s testing indicates that Muse Wonder scores higher on MemoryBench than Fable, Opus, or Astra.