Inference
EN

MiniMax M3.1 Flash vs Muse Spark 1.3

Change models
MiniMax M3.1 FlashMiniMax
Muse Spark 1.3Meta
IndexAll categories, out of 100.
69.3 ±5.4
70.1 ±4.0 best
GapIndex points below the top.
−0.8 · no clear gap
Top
CoverageIndex weight that has results.
80%
85% best
SpeedOutput tokens per second.
100/s best
76/s
EloLMArena rating and rank.
—
1491
ReleasedMonth of release.
2026-10
2026-09
Input / 1MUS dollars per 1M input tokens.
$0.15 best
$1.25
Output / 1MUS dollars per 1M output tokens.
$0.60 best
$4.25
1K chats2K tokens in, 500 out, each.
$0.60 best
$4.63
1K documents20K tokens in, 1K out, each.
$3.60 best
$29.25
1K agent tasks200K tokens in, 20K out, each.
$42.00 best
$335.00
ContextMaximum tokens in one request.
1.05M
1.05M
AgenticMulti-step tasks with tools.
71.8 best
70.4
CodingCode writing and repair.
69.4 best
68.8
ReasoningLogic problems and puzzles.
67.9
68.1 best
MultimodalTasks with images and text.
58.1
—
KnowledgeFacts and expert knowledge.
68.9 best
67.5
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
84.4
MathMath problems.
—
68.6

Benchmarks

22 shared
Only MiniMax M3.1 Flash: 14Only Muse Spark 1.3: 37
MiniMax M3.1 Flash
Muse Spark 1.3
Artificial Analysis Humanity's Last ExamKnowledge
48.0%17 / 222
48.7% best15 / 222
Artificial Analysis Long Context ReasoningReasoning
79.7%45 / 217
83.0% best18 / 217
Critical Physics TasksReasoning
22.3%25 / 214
24.9% best21 / 214
Artificial Analysis Omniscience AccuracyKnowledge
45.1% best37 / 164
43.6%41 / 164
GDPval-AA normalizedAgentic
47.9%31 / 133
59.0% best10 / 133
Vibe Code Bench v1.1Coding
73.9%45 / 123
82.9% best23 / 123
Artificial Analysis SciCodeCoding
53.4%39 / 114
58.8% best14 / 114
Legal Research BenchLegal
34.1%50 / 90
40.9% best33 / 90
Tax Agent BenchFinance
32.1%23 / 82
37.9% best14 / 82
LiveBench Data AnalysisReasoning
71.4%52 / 80
79.6% best13 / 80
LiveBench ReasoningReasoning
82.9%48 / 80
89.7% best15 / 80
Terminal-Bench 4.0Agentic
25.3% best28 / 61
10.6%45 / 61
DeepSWEAgentic
73.7%13 / 59
75.4% best6 / 59
Agent Arena steerabilityAgentic
−2.636 / 54
2.6 best22 / 54
Agent Arena task outcomeAgentic
−0.734 / 54
3.7 best20 / 54
LiveBench Coding 2.0Coding
83.9%26 / 52
90.2% best13 / 52
LiveBench DesignAgentic
74.0%31 / 52
74.8% best30 / 52
LiveBench ResearchAgentic
72.7% best29 / 52
68.6%36 / 52
LiveBench Long AgenticAgentic
34.7%29 / 51
40.4% best21 / 51
LiveBench Browser UseAgentic
92.0% best4 / 49
80.0%17 / 49