Inference
EN

Claude Sonnet 5.5 vs MiniMax M3.1 Flash

Change models
Claude Sonnet 5.5Anthropic
MiniMax M3.1 FlashMiniMax
IndexAll categories, out of 100.
77.2 ±2.6 best
69.3 ±5.4
GapIndex points below the top.
Top
−7.9
CoverageIndex weight that has results.
95% best
80%
SpeedOutput tokens per second.
107/s best
100/s
EloLMArena rating and rank.
1469
—
ReleasedMonth of release.
2026-09
2026-10
Input / 1MUS dollars per 1M input tokens.
$2.00
$0.15 best
Output / 1MUS dollars per 1M output tokens.
$10.00
$0.60 best
Batch output / 1MBatch price per 1M output tokens.
$5.00
—
1K chats2K tokens in, 500 out, each.
$9.00
$0.60 best
1K documents20K tokens in, 1K out, each.
$50.00
$3.60 best
1K agent tasks200K tokens in, 20K out, each.
$600.00
$42.00 best
ContextMaximum tokens in one request.
1M
1.05M best
AgenticMulti-step tasks with tools.
77.5 best
71.8
CodingCode writing and repair.
80.8 best
69.4
ReasoningLogic problems and puzzles.
77.8 best
67.9
MultimodalTasks with images and text.
74.4 best
58.1
KnowledgeFacts and expert knowledge.
71.3 best
68.9
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
41.5
—
MathMath problems.
76.1
—

Benchmarks

24 shared
Only Claude Sonnet 5.5: 104Only MiniMax M3.1 Flash: 12
Claude Sonnet 5.5
MiniMax M3.1 Flash
Artificial Analysis Humanity's Last ExamKnowledge
55.0% best6 / 222
48.0%17 / 222
Artificial Analysis Long Context ReasoningReasoning
82.7% best23 / 217
79.7%45 / 217
Critical Physics TasksReasoning
31.4% best7 / 214
22.3%25 / 214
Artificial Analysis Omniscience AccuracyKnowledge
51.5% best17 / 164
45.1%37 / 164
GDPval-AA normalizedAgentic
66.9% best4 / 133
47.9%31 / 133
Vibe Code Bench v1.1Coding
92.4% best2 / 123
73.9%45 / 123
Artificial Analysis SciCodeCoding
61.0% best6 / 114
53.4%39 / 114
Legal Research BenchLegal
48.1% best12 / 90
34.1%50 / 90
Tax Agent BenchFinance
42.0% best10 / 82
32.1%23 / 82
LiveBench Data AnalysisReasoning
59.5%73 / 80
71.4% best52 / 80
LiveBench ReasoningReasoning
91.6% best7 / 80
82.9%48 / 80
Terminal-Bench 4.0Agentic
64.1% best3 / 61
25.3%28 / 61
DeepSWEAgentic
71.0%21 / 59
73.7% best13 / 59
Agent Arena steerabilityAgentic
14.5 best7 / 54
−2.636 / 54
Agent Arena task outcomeAgentic
19.5 best3 / 54
−0.734 / 54
Humanity's Last Exam without toolsKnowledge
56.9% best5 / 53
43.4%20 / 53
LiveBench Coding 2.0Coding
94.7% best6 / 52
83.9%26 / 52
LiveBench DesignAgentic
85.0% best2 / 52
74.0%31 / 52
LiveBench ResearchAgentic
83.6% best10 / 52
72.7%29 / 52
LiveBench Long AgenticAgentic
78.8% best3 / 51
34.7%29 / 51