Meta has launched Muse Wonder and Muse Wonder Full, the first models from its Meta Superintelligence Labs division. These models are designed for personal and professional agentic tasks and feature Active Memory and Active Focus capabilities.
The company claims its Wonder Full model outperforms competitors in several internal benchmarks, including the Personal Superintelligence Index, SWE-Marathon coding tests, and the Harvey Legal Agent Benchmark.
In independent rankings by Artificial Analysis, Claude Opus 5.5 leads with a score of 58, while Wonder Full ties with Claude Sonnet 5.5 at 56. Muse Wonder scores 52 on the same index.
Pricing for Muse Wonder is set at $4 per million input tokens and $20 per million output tokens. This matches the pricing for Anthropic's Claude Opus 5.5 and is lower than OpenAI's rates for GPT-6 Astra.
Meta previously invested $14 billion in a deal with Scale AI to establish its Superintelligence Labs unit. The company claims the new models are its most aligned to date, reporting a zero percent rate on internal Agentic Misalignment tests.
