Inference
EN
MiniMax

MiniMax M3

MiniMax M3 is a standard model from MiniMax. 47 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Standard
  • Input: text · image · video
  • Released 2026-06
RankedEnough results for an index.
IndexAll categories, out of 100.55.7±2.9Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.79/sMedian 67/s
First tokenSeconds to the first output token.1.6 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.30batch $0.224Median $0.60
Output / 1MUS dollars per 1M output tokens.$1.20batch $0.88 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1428±4 #100Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
minimax/minimax-m3
Max outputMaximum output tokens per request.
512K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.06
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
11 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
VultrCloud$0.20$0.90524K
CoreWeaveHost$0.23$0.96262K
Deep InfraHost$0.28$1.10524K
Fireworks AIHost$0.30$1.20512K
NebiusHost$0.30$1.201.05M
Novita AIHost$0.30$1.201M
OpenRouterRouter$0.30$1.201.05M
SiliconFlowHost$0.30$1.201.05M
Together AIHost$0.30$1.20524K
GMI CloudHost$0.60$2.401.05M
SambaNovaHost$0.60$2.401.05M

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Run locally

427.04B parameters
16-bit
897.8 GB
8-bit
477.4 GB
4-bit
272.8 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.18 counted57.485 / 218
CodingCode writing and repair.12 counted57.781 / 209
ReasoningLogic problems and puzzles.5 counted50.5110 / 227
MultimodalTasks with images and text.2 counted56.550 / 84
KnowledgeFacts and expert knowledge.5 counted61.059 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted54.727 / 63
MathMath problems.3 counted41.6123 / 168

Results

47 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis IFBenchInstruction82.9%1 / 146
Artificial Analysis GPQA DiamondKnowledge92.9%13 / 199
Artificial Analysis Long Context ReasoningReasoning83.0%18 / 217
GPQA Diamond · Vals AIKnowledge92.7% ±1.414 / 135
Claw-EvalAgenticAbove fitFar above this model's index.74.5%5 / 42
Software Engineering Benchmark VerifiedCoding80.5%17 / 114
Artificial Analysis Omniscience AccuracyKnowledge44.9%38 / 164
Artificial Analysis Humanity's Last ExamKnowledge39.0%52 / 222
Artificial Analysis MMMU-ProMultimodal78.6%33 / 113
τ²-Bench Tool-Agent-User EvaluationAgentic88.9%45 / 149
Terminal-Bench 2.0 · Vals AIAgentic46.1% ±5.323 / 66
BrowseCompAgentic83.5%22 / 63
GDPval-AA normalizedAgentic37.1%48 / 133
LiveCodeBenchCoding82.2% ±1.153 / 142
GPQA diamond · Epoch AIKnowledge81.3% ±2.885 / 217
SWE-bench ProCoding59.0%36 / 83
Critical Physics TasksReasoning3.7%93 / 214
MMMU ProMultimodal81.2% ±0.941 / 92
LiveBench Data AnalysisReasoning76.2%36 / 80
MMLU ProKnowledge84.2% ±0.464 / 135
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
MCP AtlasAgentic74.2%25 / 52
Massive Multi-discipline Multimodal Understanding ProMultimodal78.1%26 / 53
SWE-benchCoding75.0% ±1.944 / 87
Artificial Analysis SciCodeCoding47.1%65 / 114
SkillsBenchCoding51.5% ±4.520 / 35
Artificial Analysis CyberGym-E2E-AAAgentic24.4%11 / 19
Vibe Code Bench v1.1Coding47.6% ±5.472 / 123
Agent Arena command recoveryAgentic0.630 / 50
OSWorld-VerifiedAgentic70.1%24 / 37
LiveBench LanguageKnowledge76.8%43 / 66
Terminal-Bench 2.1 · BenchLMAgentic66.0%57 / 86
Code MigrationCoding19.9% ±3.953 / 77
OTIS Mock AIME 2024-2025Math26.7% ±6.7138 / 200
NL2RepoCoding42.1%25 / 36
LiveBench ResearchAgentic68.0%37 / 52
Terminal-Bench 2.1 · Vals AIAgentic53.6% ±0.754 / 75
LiveBench Coding 2.0Coding70.9%40 / 52
LiveBench DesignAgentic66.2%42 / 52
ProofBench v1.1Math18.0% ±3.944 / 54
Agent Arena steerabilityAgentic−10.444 / 54
Mystery Game PuzzlesReasoning8.0% ±2.766 / 80
LiveBench ReasoningReasoning74.5%70 / 80
LiveBench Instruction FollowingInstruction57.5%59 / 66
Artificial Analysis CWE-Bench-AAAgentic41.7%17 / 19
Artificial Analysis Cyber IndexAgentic22.4%17 / 19
Terminal-Bench 4.0Agentic1.0% ±1.056 / 61
LiveBench Browser UseAgentic8.0%45 / 49
OSWorld 2.0Agentic4.6%23 / 25
LiveBench Long AgenticAgentic12.4%47 / 51
Chess PuzzlesReasoning8.0% ±2.783 / 90
Vibe Code Bench 1-100Coding9.2% ±2.924 / 26
LiveBench Agentic CodingAgentic40.7%61 / 66
Agent Arena task outcomeAgentic−18.050 / 54
OfficeQA ProMultimodal45.1%15 / 16
Artificial Analysis DeepsecBench-AAAgentic1.1%20 / 21
LiveBench CodingCoding68.2%63 / 66
LiveBench MathematicsMath76.9%64 / 66
OpenHarmony Bench v1.0Coding48.4%13 / 13
CorpFin v2Finance68.1% ±0.911 / 131
LegalBenchLegal85.4% ±0.421 / 149
MortgageTaxFinance68.4% ±0.916 / 97
MedScribeHealthcare87.3% ±2.018 / 109
Public Benefits Bench v1Public benefits60.7% ±1.33 / 13
MedCodeHealthcare46.3% ±2.137 / 106
TaxEval v2Finance72.7% ±0.958 / 143
Public Benefits Bench v1.1Public benefits64.1% ±1.224 / 50
SAGEEducation85.5% ±1.815 / 29
Legal Research BenchLegal29.8% ±3.256 / 90
Finance Agent (v2)Finance48.3% ±0.450 / 79
Tax Agent BenchFinance20.6% ±2.953 / 82
CyberBench v1.1Other65.0% ±6.134 / 48
EMBFinance47.8% ±2.957 / 75
OmniDocBench 1.5Multimodal91.6%—
Artificial Analysis Harvey LAB-AA v1.0Agentic88.4%—
United States of America Mathematical Olympiad 2026Math85.7%—
Video-MME with subtitleMultimodal85.4%—
VideoMMMUMultimodal84.6%—
BankerToolBenchAgentic76.1%—
GDPval rubricsAgentic74.7%—
SVG-BenchCoding63.7%—
VIBE V2Coding50.1%—
KernelBench HardCoding28.8%—
ResearchClawBenchAgentic19.8%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEpoch AICC BYLiveBenchNo licence statedLMArenaCC BY 4.0models.devMIT

History

5 releases
Index55.8 → 55.7
23 Oct 202612 Nov 2026
Output / 1M$1.20 → $1.20
23 Oct 202612 Nov 2026

Questions

How much does MiniMax M3 cost?
$0.30 per 1M input tokens, $1.20 per 1M output tokens.
How good is MiniMax M3?
Its index is 55.7 out of 100. That is place 96 of 283 ranked models.
What is MiniMax M3 best at?
Knowledge: 61.0 out of 100, place 59 of 268.
How fast is MiniMax M3?
79 output tokens per second, 1.6 s to the first token.
What is the context window of MiniMax M3?
1.05M tokens.
Is MiniMax M3 open source?
Yes. Its weights are published.
When was MiniMax M3 released?
June 2026.

Badge

MiniMax M3 on Inference360: place 96 of 283, index 55.7
HTML
<a href="https://inference360.ai/models/minimax-m3"><img src="https://inference360.ai/badges/minimax-m3.svg" alt="MiniMax M3 on Inference360: place 96 of 283, index 55.7" height="20"></a>
Markdown
[![MiniMax M3 on Inference360: place 96 of 283, index 55.7](https://inference360.ai/badges/minimax-m3.svg)](https://inference360.ai/models/minimax-m3)

Compare with

Apodex 1.1 Mini55.7Grok 4.20 (Reasoning)55.5GLM-5.156.0Claude Opus 4.5 Thinking56.3MiniMax M2.751.2

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeGLM-5.3-Flash63.1 · $0.078/1M

More from MiniMax

All models
MiniMax M2.751.2MiniMax M2.550.2MiniMax M3.1 Flash PreviewUnrankedMiniMax M2-herUnrankedMiniMax M3.1 Flash69.3