Inference
EN
Meta

Muse Wonder

Muse Wonder is a reasoning model from Meta. 43 benchmarks count toward its index, in 6 categories.

  • Closed weights
  • Reasoning
  • Input: text · image · video · file · audio
  • Released 2026-10
RankedEnough results for an index.
IndexAll categories, out of 100.73.1±4.0Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.80/sMedian 67/s
First tokenSeconds to the first output token.2.7 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$4.00Median $0.60
Output / 1MUS dollars per 1M output tokens.$20.00Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1503±6 #6Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
meta/muse-wonder
Max outputMaximum output tokens per request.
944K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.20
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.14 counted73.013 / 218
CodingCode writing and repair.12 counted72.315 / 209
ReasoningLogic problems and puzzles.7 counted73.119 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.4 counted68.626 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted74.4—
MathMath problems.5 counted73.517 / 168

Results

43 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
DeepSWEAgentic78.8%2 / 59
GDPval-AA normalizedAgentic63.5%5 / 133
LiveBench ResearchAgentic88.0%2 / 52
Artificial Analysis Humanity's Last ExamKnowledge50.1%12 / 222
JobBenchAgentic70.5%2 / 36
LiveBench Data AnalysisReasoning81.6%5 / 80
Critical Physics TasksReasoningAbove fitFar above this model's index.28.3%15 / 214
LiveBench MathematicsMath96.5%5 / 66
Code MigrationCoding62.6% ±4.46 / 77
OTIS Mock AIME 2024-2025Math99.4% ±0.417 / 200
Artificial Analysis SciCodeCoding60.3%10 / 114
LiveBench Agentic CodingAgentic62.7%9 / 66
LiveBench ReasoningReasoning90.6%11 / 80
Artificial Analysis Long Context ReasoningReasoning80.7%37 / 217
LiveBench Long AgenticAgentic53.2%9 / 51
Vibe Code Bench v1.1Coding83.1% ±2.522 / 123
Terminal-Bench 4.0Agentic49.5% ±2.211 / 61
CWE-bench v2Agentic66.1%2 / 11
Mystery Game PuzzlesReasoning51.0% ±5.015 / 80
FrontierMath-Tiers-1-3-v2-PrivateMath82.1% ±2.318 / 90
Artificial Analysis GDP.pdfAgentic30.6%5 / 22
Vibe Code Bench 1-100Coding26.6% ±4.36 / 26
FrontierMath-Tier-4-v2-PrivateMath70.7% ±7.217 / 72
Agent Arena task outcomeAgentic5.413 / 54
LiveBench CodingCoding80.7%16 / 66
LiveBench Instruction FollowingInstruction73.0%16 / 66
ProofBench v1.1Math84.0% ±3.715 / 54
LiveBench LanguageKnowledge82.9%19 / 66
IOICoding66.9% ±4.915 / 45
cursorBench40Coding50.4%7 / 21
Furniture AssemblyReasoning54.2% ±6.413 / 37
Artificial Analysis Omniscience AccuracyKnowledge38.0%59 / 164
LiveBench Coding 2.0Coding86.2%20 / 52
Agent Arena steerabilityAgentic2.622 / 54
LiveBench Browser UseAgentic72.0%20 / 49
Multimodal Knowledge Agent BenchmarkAgentic47.5%8 / 19
Artificial Analysis Terminal-Bench v4.0Agentic52.0%10 / 22
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
EBR-benchReasoning31.4% ±4.215 / 31
MirrorCodeCoding45.6% ±10.620 / 39
Bug Hunt BenchCoding31.7%13 / 25
Medical Long Context Reasoning (MLCR-AA)Reasoning22.2%12 / 22
HealthBench ProfessionalKnowledge60.5%10 / 18
FrontierCode 2.0Coding24.3%15 / 27
LiveBench DesignAgentic75.1%29 / 52
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding50.0%11 / 19
OSWorld 2.1 (all 108 tasks)Agentic72.6%9 / 11
Artificial Analysis AutomationBenchAgentic39.1%22 / 23
CyberBench v1.1Other80.4% ±5.11 / 48
Legal Research BenchLegal50.5% ±3.57 / 90
HLAB-Vals-1.1Legal23.8% ±3.64 / 35
Public Benefits Bench v1.1Public benefits70.2% ±1.26 / 50
Tax Agent BenchFinance36.3% ±3.515 / 82
Finance Agent (v2)Finance56.6% ±0.418 / 79
EMBFinance67.1% ±2.918 / 75
Terminal-Bench-ScienceScience22.9% ±5.111 / 44
MysteryMechanismScience27.5% ±3.016 / 30
BioMysteryBenchScience67.0% ±0.420 / 30
CUA-benchOther7.2%6 / 9
SWE-MarathonCoding53.1%—
Harvey LAB-AA v1.1Agentic10.0%—
STEMBench120: six-domain scientific research progressReasoning2.7%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYLiveBenchNo licence statedVals AINo licence statedLMArenaCC BY 4.0

History

5 releases
Index74.6 → 73.1
23 Oct 202612 Nov 2026
Output / 1M$20.00 → $20.00
23 Oct 202612 Nov 2026

Questions

How much does Muse Wonder cost?
$4.00 per 1M input tokens, $20.00 per 1M output tokens.
How good is Muse Wonder?
Its index is 73.1 out of 100. That is place 16 of 283 ranked models.
What is Muse Wonder best at?
Agentic: 73.0 out of 100, place 13 of 218.
How fast is Muse Wonder?
80 output tokens per second, 2.7 s to the first token.
What is the context window of Muse Wonder?
1.05M tokens.
Is Muse Wonder open source?
No. Its weights are not published.
When was Muse Wonder released?
October 2026.

Badge

Muse Wonder on Inference360: place 16 of 283, index 73.1
HTML
<a href="https://inference360.ai/models/muse-wonder"><img src="https://inference360.ai/badges/muse-wonder.svg" alt="Muse Wonder on Inference360: place 16 of 283, index 73.1" height="20"></a>
Markdown
[![Muse Wonder on Inference360: place 16 of 283, index 73.1](https://inference360.ai/badges/muse-wonder.svg)](https://inference360.ai/models/muse-wonder)

Compare with

GPT-5.4 Pro73.3GPT-5.6 Sol73.0GPT-6 Sol72.8Claude Opus 573.8Claude Fable 574.0

Lower price

MiMo-V2.6-Pro70.2 · $0.87/1MHy472.0 · $1.25/1MGPT-6.1 Terra77.1 · $5.00/1M

More from Meta

All models
Muse Wonder Full77.3Muse Spark Open65.1Muse Ember64.8Muse Spark 1.3 ContributorUnrankedMuse Spark 1.370.1