MMuse Wonder 1.0
muse family · launched 2026-10-12
MultimodalReasoningToolsStructuredMuse Wonder is a frontier multimodal reasoning model from Meta for personal and professional agents, long-horizon coding, and computer use. It combines Active Memory, which carries relevant context across conversations, with Active Focus, which keeps it working through blockers toward a goal.
What this model supports
Strengths by domain
Every result has a source
Raw results and standardized readings are shown together. Missing tests add neither points nor penalties.
Coding1 results
LiveBench 2.0 · CodingLiveBench86.20 scoreReading 86.2
Imported from LiveBench 2.0. The published 0–100 score is used directly; it is not z-score normalized.
Configuration: LiveBench 2.0Open evidenceResearch & reasoning1 results
LiveBench 2.0 · ResearchLiveBench90.60 scoreReading 90.6
Imported from LiveBench 2.0. The published 0–100 score is used directly; it is not z-score normalized.
Configuration: LiveBench 2.0Open evidenceProfessional work1 results
LiveBench 2.0 · ProfessionalLiveBench70.48 scoreReading 70.5
Imported from LiveBench 2.0. The published 0–100 score is used directly; it is not z-score normalized. Professional = long-agentic, browser, data-analysis and design average.
Configuration: LiveBench 2.0Open evidenceKnowledge & instruction1 results
LiveBench 2.0 · KnowledgeLiveBench88.00 scoreReading 88.0
Imported from LiveBench 2.0. The published 0–100 score is used directly; it is not z-score normalized. Knowledge uses the published LiveBench research suite.
Configuration: LiveBench 2.0Open evidenceHuman preference & indices2 results
LiveBenchLiveBench78.10 scoreReading 78.1
Imported from LiveBench 2.0. The published 0–100 score is used directly; it is not z-score normalized.
Configuration: LiveBench 2.0Open evidenceAA IndexArtificial Analysis78.10 indexReading 76.2
Imported from Artificial Analysis. The published 0–100 score is used directly; it is not z-score normalized.
Configuration: LiveBench 2.0Open evidence