Model
Model explorer

Muse Wonder Full

CLOSED
Meta · Muse Wonder family · released Oct 12, 2026

The larger Muse Wonder variant, released October 12, 2026 but not offered in the Meta Model API: it is available only inside Muse Code on the $50/month Power Usage and $200/month Maximum Usage plans (dev.meta.ai lists no per-token price), so price is null and apiAvailable false, following the Gemini 3.8 Flash Cyber precedent (Gemini 4 Argon followed it too until its Oct 13 GA). Meta's scorecard and Artificial Analysis (Intelligence Index 56, #4) put it just behind Claude Opus 5.5 (58). It leads Meta's table on SWE-Marathon (57.5, run at xhigh), DeepSWE 1.1 (79.0) and Vals Finance Agent v2; Vals AI independently ranks it #5 of 47 on the Vals Index (64.57%) and #4 of 79 on Harvey's Legal Agent Benchmark. Cyber rows come from the model card's risk section (ExploitGym's 92 flags are not filed: only 21 used the intended vulnerability). NOT FILED: Personal Superintelligence Index (59.1) and MemoryBench (85.5) are Meta-internal; Harvey LAB-AA 13.1 and AutomationBench-AA 45.4 are Artificial Analysis variants that do not match the corpus columns; ProgramBench 59.0 conflicts with Vals' 0.50% on its own ProgramBench metric.

ReasoningCodingVisionFunction callingTool useAgentic
3100.5
Elo · rank #3
Parameters
Undisclosed
Active params
—
Context
1M tokens
Architecture
Proprietary multimodal reasoning model (architecture undisclosed); Meta's most capable Muse Wonder variant, 1M-token context
License
Proprietary
Languages
—
API price (in/out)
No hosted API
Modalities
text · vision · audio · video
Benchmark results
Bar shows position within the tracked field; marker = field best
CyberGymCoding88.4%#3
best: MiMo-V2.6-Flash · 95.1%
DeepSWECoding79.0%#1
best: this model · 79.0%
ExploitBenchAgents58.9%#6
best: GPT-6 Astra · 100.0%
GDP.PDFVision33.2%#5
best: GPT-5.6 Sol · 40.0%
best: Muse Spark 1.2 · 25.4%
HealthBench ProfessionalKnowledge62.3%#5
best: Claude Opus 5.5 · 65.6%
JobBenchAgents75.3%#1
best: this model · 75.3%
OSWorld 2.0Agents78.4%#3
best: Claude Opus 5.5 · 81.8%
SWE-MarathonCoding57.5%#1
best: this model · 57.5%
Terminal-Bench 4.0Agents58.1%#6
best: Claude Sonnet 5.5 · 70.6%
best: GPT-6 Astra · 62.9%
Vals Finance AgentAgents59.1%#5
best: Gemini 4 Argon · 65.4%
Run it locally
Closed weights — available via API only. No local deployment.
Input / M tok
—
Output / M tok
—
Muse Wonder family
Elo progression across releases
API price — · each benchmark row carries its own source badge (see methodology)