Inference
EN

Muse Spark 1.3 vs Step 5 Preview

Change models
Muse Spark 1.3Meta
Step 5 PreviewStepFun
IndexAll categories, out of 100.
70.1 ±4.0 best
68.4 ±3.7
GapIndex points below the top.
Top
−1.7 · no clear gap
CoverageIndex weight that has results.
85%
90% best
SpeedOutput tokens per second.
76/s
105/s best
EloLMArena rating and rank.
1491 best
1443
ReleasedMonth of release.
2026-09
2026-09
Input / 1MUS dollars per 1M input tokens.
$1.25
$1.00 best
Output / 1MUS dollars per 1M output tokens.
$4.25
$2.70 best
1K chats2K tokens in, 500 out, each.
$4.63
$3.35 best
1K documents20K tokens in, 1K out, each.
$29.25
$22.70 best
1K agent tasks200K tokens in, 20K out, each.
$335.00
$254.00 best
ContextMaximum tokens in one request.
1.05M best
1M
AgenticMulti-step tasks with tools.
70.4 best
68.8
CodingCode writing and repair.
68.8
71.4 best
ReasoningLogic problems and puzzles.
68.1 best
67.4
MultimodalTasks with images and text.
—
62.2
KnowledgeFacts and expert knowledge.
67.5 best
63.5
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
84.4
—
MathMath problems.
68.6 best
61.7

Benchmarks

30 shared
Only Muse Spark 1.3: 29Only Step 5 Preview: 32
Muse Spark 1.3
Step 5 Preview
Artificial Analysis Humanity's Last ExamKnowledge
48.7% best15 / 222
46.5%23 / 222
Artificial Analysis Long Context ReasoningReasoning
83.0%18 / 217
88.3% best2 / 217
Critical Physics TasksReasoning
24.9% best21 / 214
20.9%26 / 214
Artificial Analysis Omniscience AccuracyKnowledge
43.6% best41 / 164
36.4%64 / 164
GDPval-AA normalizedAgentic
59.0% best10 / 133
54.3%24 / 133
Vibe Code Bench v1.1Coding
82.9% best23 / 123
70.4%48 / 123
Artificial Analysis SciCodeCoding
58.8%14 / 114
58.9% best13 / 114
Legal Research BenchLegal
40.9% best33 / 90
40.4%35 / 90
Terminal-Bench 2.1 · BenchLMAgentic
88.8% best7 / 86
85.0%20 / 86
Tax Agent BenchFinance
37.9% best14 / 82
27.3%39 / 82
LiveBench Data AnalysisReasoning
79.6% best13 / 80
63.5%69 / 80
LiveBench ReasoningReasoning
89.7% best15 / 80
83.8%44 / 80
Finance Agent (v2)Finance
58.9% best8 / 79
50.7%42 / 79
Code MigrationCoding
27.6%44 / 77
39.0% best31 / 77
EMBFinance
62.7% best31 / 75
57.1%41 / 75
Terminal-Bench 4.0Agentic
10.6%45 / 61
31.8% best23 / 61
DeepSWEAgentic
75.4% best6 / 59
67.7%27 / 59
Agent Arena steerabilityAgentic
2.6 best22 / 54
−0.930 / 54
Agent Arena task outcomeAgentic
3.7 best20 / 54
−1.535 / 54
ProofBench v1.1Math
55.0% best30 / 54
42.0%37 / 54