Muse Wonder

Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index v4.3 at maximum reasoning, placing it second behind Anthropic's Claude Opus 5.5, while standard Muse Wonder reached 52 and predecessor Muse Spark 1.3 sat at 48. CryptoBriefing reports Muse Wonder Full landed in the top three on six of ten sub-evaluations, with strongest showings on agentic and coding benchmarks, and claims it surpasses GPT-6 Astra and Claude Fable 5.1 on coding, legal and automation tests. Pricing on the Meta Model API is set at $4 per million input tokens and $20 per million output tokens, matching Claude Opus 5.5, according to the same report. Distribution is split by tier: standard Muse Wonder is on the Meta Model API and Muse Code, while Full remains exclusive to Muse Code subscribers, with the same Active Memory and Active Focus features described in the launch materials.

Meta released Muse Wonder and a larger sibling, Muse Wonder Full, on October 12, 2026, calling them the company's most capable frontier-class models for personal and professional agentic tasks. According to Meta AI Research, both models were trained at Meta's gigawatt-scale Prometheus data center and introduce two design pillars: Active Memory, which carries relevant context across conversations, and Active Focus, which keeps agents working through long-running tasks. Per Meta's blog, Muse Wonder is available today through the Meta Model API and Muse Code, while Muse Wonder Full is restricted to Muse Code subscribers at maximum reasoning effort. Meta says the series achieves state-of-the-art results across benchmarks for agents, memory and coding, and that Wonder will later power experiences in the Muse app starting with subscribers, plus future specialized model variants built on the same architecture.

CellCog's row-by-row analysis of Meta's own benchmark table says the Wonder family leads 7 of 14 published benchmarks, Claude Opus 5.5 leads 6, and GPT-6 Astra leads 1. The Wonder family is strongest on long-running work including SWE-Marathon, DeepSWE, MemoryBench and AutomationBench, while Opus 5.5 leads on Terminal-Bench, OSWorld and ProgramBench and tops Meta's own Artificial Analysis summary at 58 to 56. CellCog lists Muse Wonder at $4 per million input tokens, $0.20 cached, and $20 per million output on the Meta Model API with a 1M-token context window, with Wonder Full reserved for Muse Code subscribers. The piece also notes Meta's positioning claim that plain Muse Wonder beats Opus 5.5 on 5 of the 14 rows, framing the launch as a tighter race than Meta's headline language suggests.