Inference
EN

Muse Spark vs Qwen3.8-27B

Change models
Muse SparkMeta
Qwen3.8-27BAlibaba
IndexAll categories, out of 100.
59.8 ±3.5 best
59.2 ±2.1
GapIndex points below the top.
Top
−0.6 · no clear gap
CoverageIndex weight that has results.
95%
100% best
SpeedOutput tokens per second.
—
113/s
EloLMArena rating and rank.
1474 best
1436
ReleasedMonth of release.
2026-04
2026-08
Input / 1MUS dollars per 1M input tokens.
—
$0.15
Output / 1MUS dollars per 1M output tokens.
—
$1.88
1K chats2K tokens in, 500 out, each.
—
$1.24
1K documents20K tokens in, 1K out, each.
—
$4.88
1K agent tasks200K tokens in, 20K out, each.
—
$67.50
ContextMaximum tokens in one request.
262K
1.05M best
AgenticMulti-step tasks with tools.
55.9
57.4 best
CodingCode writing and repair.
53.3
60.3 best
ReasoningLogic problems and puzzles.
59.8
61.2 best
MultimodalTasks with images and text.
63.0
63.6 best
KnowledgeFacts and expert knowledge.
67.8 best
51.2
MultilingualTasks in many languages.
—
60.6
InstructionTasks with strict prompt rules.
60.0
68.5 best
MathMath problems.
60.2 best
56.3

Benchmarks

22 shared
Only Muse Spark: 25Only Qwen3.8-27B: 78
Muse Spark
Qwen3.8-27B
Artificial Analysis Humanity's Last ExamKnowledge
40.7% best45 / 222
33.9%67 / 222
Artificial Analysis Long Context ReasoningReasoning
78.0%70 / 217
82.0% best29 / 217
Critical Physics TasksReasoning
11.3% best60 / 214
5.4%81 / 214
Artificial Analysis GPQA DiamondKnowledge
88.4%51 / 199
90.5% best32 / 199
Artificial Analysis Omniscience AccuracyKnowledge
49.6% best21 / 164
17.3%143 / 164
LegalBenchLegal
84.2% best37 / 149
82.4%72 / 149
TaxEval v2Finance
77.7% best3 / 143
70.8%86 / 143
GPQA Diamond · Vals AIKnowledge
89.6% best28 / 135
88.9%32 / 135
MMLU ProKnowledge
87.3% best29 / 135
84.3%63 / 135
GDPval-AA normalizedAgentic
24.9%73 / 133
45.9% best38 / 133
Vibe Code Bench v1.1Coding
19.7%92 / 123
64.8% best54 / 123
Artificial Analysis MMMU-ProMultimodal
80.5% best18 / 113
76.3%46 / 113
MedScribeHealthcare
85.9% best25 / 109
83.8%41 / 109
MedCodeHealthcare
51.3% best16 / 106
28.7%98 / 106
MMMU ProMultimodal
87.4% best16 / 92
83.9%28 / 92
SWE-benchCoding
74.4%48 / 87
86.0% best15 / 87
GPQA Diamond · BenchLMKnowledge
89.5% best33 / 84
89.2%37 / 84
SWE-bench ProCoding
52.4%64 / 83
61.7% best27 / 83
Humanity's Last ExamKnowledge
50.4% best21 / 77
30.8%45 / 77
Humanity's Last Exam without toolsKnowledge
42.8% best24 / 53
30.8%39 / 53