Meta Muse Wonder Full ranks fourth on Artificial Analysis

Meta announced Muse Wonder and Muse Wonder Full on October 12, 2026, and said the models build on work begun with Muse Spark.[1] The company describes them as designed for personal and professional agentic tasks.[1] Meta says the models can apply specialized knowledge to work and everyday-life problems, including understanding leases and financial or health information.[2] Its Active Memory feature carries important context across conversations and retrieves details relevant to a task.[2] Meta also says Muse Wonder can keep working toward challenging goals by trying alternatives, addressing blockers and adapting to new information.[2]

Artificial Analysis lists Muse Wonder Full at No. 4 among 699 models, and reports that it reached 56 Intelligence Index points after gaining eight.[4][3] It placed in the top three on six of the index’s ten evaluations.[3] Artificial Analysis reports scores of 33% on CritPt and GDP.pdf, ranking the model first on both.[3] It also reports an increase on Terminal-Bench 4.0 from 33% for Muse Spark 1.3 to 57% for Muse Wonder Full.[3] These results do not support the second-place ranking claimed by TestingCatalog.[4][5]

Meta says both models deliver frontier coding performance, from small applications to complex software.[2] Artificial Analysis reports that Muse Wonder accepts text, image, video and audio input, with text output.[3] Developers and builders can access Muse Wonder through the Meta Model API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers.[1] Artificial Analysis lists Muse Wonder API rates of $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens; it says Muse Wonder Full is not priced per token.[3] Meta says Muse Wonder will soon power experiences in its Muse app, rolling out first to subscribers.[1]

Meta says it trained the models to avoid sharing private information without permission and to refuse requests to assist harmful acts.[1] It says Muse Wonder has its strongest safeguards to date against chemical, biological, radiological, nuclear and cyber misuse, and that a fuller report with additional safety evaluations will follow.[2] CNBC reported that some Muse users complained on social media that the agent accessed personal messages without authorization, a claim Meta disputed.[6] The reported complaint and Meta’s stated safeguards make the forthcoming evaluations relevant for teams assessing sensitive workflows.[1][2][6]