SShogo Bench

Muse Wonder

Muse 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.

Shogo Agent Index

Estimated#15 of 92
60.9range 51.9–69.9 · 22% of weight measured
Agentic coding
68
11% measured
Long-horizon autonomy
70
50% measured
Knowledge-work agents
50
32% measured
Computer use
50
estimated
Security & context learning
57
estimated

Price

Input
$4.00/M
Output
$20.00/M
Cache read
$0.20/M
3:1 blend
$8.00/M
Agent-effective
$1.69/M

Cheapest host for agents: Google at $1.69/M.

Agent-effective = 20 input tokens per output token, 85% cache hits. Why.

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 84
Terminal-Bench 2.0est.8%predicted 89
DeepSWEest.6%predicted 67
FrontierSWEest.5%predicted 36
CursorBench50.4%4%source · $5.25/task
GSO-Benchest.4%predicted 54
METR Time Horizonsest.10%predicted 82
Vending-Bench 2$9,4326%source
PostTrainBench24.2%4%OpenCode
APEX-Agents66.1%8%source
GDPvalest.7%predicted 60
Remote Labor Indexest.5%predicted 10
DeepResearch Benchest.5%predicted 51
OSWorld 2.0est.6%predicted 23
OSWorldest.4%predicted 89
Cybenchest.5%predicted 87
CL-benchest.5%predicted 26

Other benchmarks

BenchmarkCategoryScoreSource
CritPtReasoning28.3%source
DTBenchKnowledge95.2%source
FrontierMath Tiers 1-3Math82.1%source
FrontierMath Tier 4Math70.7%source
LMCAKnowledge56.8%source
MirrorCodeCoding45.6%source
OTIS Mock AIMEMath99.4%source
ProofBenchMath84.0%source
SciCodeCoding60.3%source
WeirdML v3Coding15.1%codex_cli

API hosts

Live prices and uptime for every host serving Muse Wonder via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
Google$4.00$20.00$0.20$1.69100.0%—
Meta$4.00$20.00$0.20$1.6999.9%—