Inference
EN

Muse Spark 1.3 vs Muse Wonder Full

Change models
Muse Spark 1.3Meta
Muse Wonder FullMeta
IndexAll categories, out of 100.
70.1 ±4.0
77.3 ±5.3 best
GapIndex points below the top.
−7.2
Top
CoverageIndex weight that has results.
85% best
80%
SpeedOutput tokens per second.
76/s
—
EloLMArena rating and rank.
1491
—
ReleasedMonth of release.
2026-09
2026-10
Input / 1MUS dollars per 1M input tokens.
$1.25
—
Output / 1MUS dollars per 1M output tokens.
$4.25
—
1K chats2K tokens in, 500 out, each.
$4.63
—
1K documents20K tokens in, 1K out, each.
$29.25
—
1K agent tasks200K tokens in, 20K out, each.
$335.00
—
ContextMaximum tokens in one request.
1.05M best
1M
AgenticMulti-step tasks with tools.
70.4
78.2 best
CodingCode writing and repair.
68.8
77.3 best
ReasoningLogic problems and puzzles.
68.1
75.6 best
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
67.5
71.7 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
84.4
—
MathMath problems.
68.6
79.6 best

Benchmarks

24 shared
Only Muse Spark 1.3: 35Only Muse Wonder Full: 18
Muse Spark 1.3
Muse Wonder Full
Artificial Analysis Humanity's Last ExamKnowledge
48.7%15 / 222
53.6% best9 / 222
Artificial Analysis Long Context ReasoningReasoning
83.0%18 / 217
83.7% best10 / 217
Critical Physics TasksReasoning
24.9%21 / 214
32.9% best2 / 214
Artificial Analysis Omniscience AccuracyKnowledge
43.6% best41 / 164
41.0%46 / 164
GDPval-AA normalizedAgentic
59.0%10 / 133
67.9% best2 / 133
Vibe Code Bench v1.1Coding
82.9%23 / 123
86.0% best14 / 123
Artificial Analysis SciCodeCoding
58.8%14 / 114
62.6% best4 / 114
Legal Research BenchLegal
40.9%33 / 90
53.4% best6 / 90
Tax Agent BenchFinance
37.9%14 / 82
43.0% best8 / 82
Finance Agent (v2)Finance
58.9%8 / 79
59.1% best6 / 79
Code MigrationCoding
27.6%44 / 77
61.0% best7 / 77
EMBFinance
62.7%31 / 75
71.9% best10 / 75
ProgramBenchCoding
0.5%19 / 65
0.5%19 / 65
Terminal-Bench 4.0Agentic
10.6%45 / 61
58.1% best6 / 61
DeepSWEAgentic
75.4%6 / 59
79.0% best1 / 59
Agent Arena steerabilityAgentic
2.622 / 54
9.6 best12 / 54
Agent Arena task outcomeAgentic
3.720 / 54
11.1 best9 / 54
ProofBench v1.1Math
55.0%30 / 54
94.0% best11 / 54
CyberBench v1.1Other
69.4%27 / 48
73.6% best11 / 48
MazeBench: visual spatial reasoning in a 3D open worldAgentic
0.0%23 / 48
17.0% best2 / 48