SShogo Bench

Muse Spark 1.3

Muse Spark 1.3 is a multimodal reasoning model from Meta for long-running agentic, multi-agent, and coding workflows. It improves long-horizon agent collaboration, instruction following, and coding efficiency relative to Muse Spark 1.2.

Shogo Agent Index

Estimated#28 of 92
54.7range 41.6–67.7 · 17% of weight measured
Agentic coding
60
26% measured
Long-horizon autonomy
64
estimated
Knowledge-work agents
45
32% measured
Computer use
43
estimated
Security & context learning
53
estimated

Price

Input
$0.10/M
Output
$0.20/M
Cache read
$0.002/M
3:1 blend
$0.13/M
Agent-effective
$0.030/M

Cheapest host for agents: Google at $0.50/M.

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

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 81
Terminal-Bench 2.0est.8%predicted 82
DeepSWEest.6%predicted 58
FrontierSWE25.8%5%proximus · $243.34/task
CursorBench41.6%4%source · $2.64/task
GSO-Benchest.4%predicted 41
METR Time Horizonsest.10%predicted 79
Vending-Bench 2est.6%predicted 64
PostTrainBenchest.4%predicted 28
APEX-Agents58.6%8%source
GDPvalest.7%predicted 53
Remote Labor Indexest.5%predicted 6
DeepResearch Benchest.5%predicted 49
OSWorld 2.0est.6%predicted 15
OSWorldest.4%predicted 84
Cybenchest.5%predicted 82
CL-benchest.5%predicted 24

Other benchmarks

BenchmarkCategoryScoreSource
Chess PuzzlesReasoning37.0%source
CritPtReasoning26.0%source
DTBenchKnowledge96.5%source
FrontierMath Tiers 1-3Math74.4%source
FrontierMath Tier 4Math46.3%source
GeoBenchMultimodal86.0%source
LMCAKnowledge53.9%source
MirrorCodeCoding54.4%source
OTIS Mock AIMEMath99.2%source
ProofBenchMath58.0%source
SciCodeCoding59.7%source
SimpleBenchReasoning81.8%source
WebDev ArenaCoding1652source
WeirdML v3Coding6.7%codex_cli

API hosts

Live prices and uptime for every host serving Muse Spark 1.3 via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
Google$1.25$4.25$0.15$0.5099.4%—
Meta$1.25$4.25$0.15$0.5099.8%—