Inference
EN
MiniMax

MiniMax M2.7

MiniMax M2.7 is a standard model from MiniMax. 30 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Standard
  • Input: text
  • Released 2026-03
RankedEnough results for an index.
IndexAll categories, out of 100.51.2±4.5Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.35/sMedian 67/s
First tokenSeconds to the first output token.1.2 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.21Median $0.60
Output / 1MUS dollars per 1M output tokens.$0.84Median $2.55
ContextMaximum tokens in one request.197KMedian 262K
EloLMArena rating and rank.1405±3 #151Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
minimax/minimax-m2.7
Max outputMaximum output tokens per request.
177K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
No
Cache read / 1MUS dollars per 1M cached tokens.
$0.042
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.
5 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
OpenRouterRouter$0.21$0.84197K
Deep InfraDeprecatedHost$0.25$1.00197K
GMI CloudHost$0.30$1.20197K
Novita AIHost$0.30$1.20205K
Together AIDeprecatedHost$0.30$1.20197K

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

Run locally

228.69B parameters
16-bit
481.2 GB
8-bit
256.1 GB
4-bit
146.6 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.10 counted52.0118 / 218
CodingCode writing and repair.9 counted52.3124 / 209
ReasoningLogic problems and puzzles.2 counted49.5120 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.4 counted50.2133 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted59.8—
MathMath problems.4 counted47.794 / 168

Results

30 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis IFBenchInstruction75.7%19 / 146
AIMEMath91.0% ±0.726 / 95
Artificial Analysis GPQA DiamondKnowledge87.4%57 / 199
Artificial Analysis Long Context ReasoningReasoning78.3%63 / 217
GPQA Diamond · Vals AIKnowledge86.6% ±2.041 / 135
τ³-Bench Tool-Agent-User EvaluationAgentic67.6%10 / 32
Terminal-Bench 2.0 · Vals AIAgentic47.2% ±5.322 / 66
Artificial Analysis Humanity's Last ExamKnowledge29.6%82 / 222
τ²-Bench Tool-Agent-User EvaluationAgentic84.8%58 / 149
SWE-bench Verified (mini-swe-agent-v2)Coding75.4%2 / 5
Artificial Analysis SciCodeCoding50.1%52 / 114
LiveCodeBenchCoding79.9% ±1.171 / 142
Terminal-Bench 2.0 · BenchLMAgentic57.0%22 / 41
GDPval-AA normalizedAgentic25.5%72 / 133
Artificial Analysis Omniscience AccuracyKnowledge26.8%92 / 164
GPQA Diamond · BenchLMKnowledge87.0%49 / 84
SWE-benchCoding73.8% ±2.051 / 87
SWE-bench ProCoding56.2%50 / 83
Critical Physics TasksReasoning0.6%132 / 214
ToolathlonAgentic46.3%14 / 22
MMLU ProKnowledge80.4% ±0.488 / 135
Humanity's Last ExamKnowledge28.0%53 / 77
Gert Labs Composite Game BenchmarkAgentic40.4%36 / 52
IOI v1Coding4.9% ±2.444 / 61
NL2RepoCoding39.8%27 / 36
AIME 2026Math89.8%26 / 34
Terminal-Bench 2.1 · Vals AIAgentic48.7% ±3.162 / 75
MMLU-Pro first-party comparison snapshotKnowledge80.8%5 / 6
Vibe Code Bench v1.1Coding11.9% ±2.9103 / 123
Code MigrationCoding8.7% ±1.865 / 77
Harvard-MIT Mathematics Tournament February 2026Math72.7%23 / 27
APEX-Agents-AAAgentic10.6%27 / 30
Harvard-MIT Mathematics Tournament November 2025Math81.0%11 / 12
Claw-EvalAgentic48.7%39 / 42
React Native EvalsCoding71.4%15 / 16
LegalBenchLegal84.0% ±0.443 / 149
CorpFin v2Finance61.2% ±1.055 / 131
CaseLaw v2Legal60.9% ±0.823 / 53
ProgramBenchCoding0.0%34 / 65
MedScribeHealthcare79.9% ±1.960 / 109
Finance Agent v1.1Finance48.4% ±2.829 / 51
MedCodeHealthcare34.4% ±2.079 / 106
TaxEval v2Finance66.6% ±0.9110 / 143
EMBFinance28.5% ±2.465 / 75
Legal Research BenchLegal10.6% ±2.179 / 90
Tax Agent BenchFinance6.9% ±1.775 / 82
Finance Agent (v2)Finance27.9% ±0.574 / 79
AIME25 first-party comparison snapshotMath80.0%—
MLE-Bench LiteAgentic66.6%—
MM-ClawBenchAgentic62.7%—
VIBE-ProCoding55.6%—
Multi-SWE BenchCoding52.7%—
SWE-RebenchCoding51.9%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedmodels.devMIT

History

5 releases
Index52.1 → 51.2
23 Oct 202612 Nov 2026
Output / 1M$1.20 → $0.84
23 Oct 202612 Nov 2026

Questions

How much does MiniMax M2.7 cost?
$0.21 per 1M input tokens, $0.84 per 1M output tokens.
How good is MiniMax M2.7?
Its index is 51.2 out of 100. That is place 141 of 283 ranked models.
What is MiniMax M2.7 best at?
Knowledge: 50.2 out of 100, place 133 of 268.
How fast is MiniMax M2.7?
35 output tokens per second, 1.2 s to the first token.
What is the context window of MiniMax M2.7?
197K tokens.
Is MiniMax M2.7 open source?
Yes. Its weights are published.
When was MiniMax M2.7 released?
March 2026.

Badge

MiniMax M2.7 on Inference360: place 141 of 283, index 51.2
HTML
<a href="https://inference360.ai/models/minimax-m2-7"><img src="https://inference360.ai/badges/minimax-m2-7.svg" alt="MiniMax M2.7 on Inference360: place 141 of 283, index 51.2" height="20"></a>
Markdown
[![MiniMax M2.7 on Inference360: place 141 of 283, index 51.2](https://inference360.ai/badges/minimax-m2-7.svg)](https://inference360.ai/models/minimax-m2-7)

Compare with

Qwen3.5-27B51.3Nemotron 3.5 Super VL 120B A12B51.0GPT-5.4 nano50.9GLM-551.6MiniMax M2.550.2

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeNemotron 3 Ultra50.8 · Free

More from MiniMax

All models
MiniMax M2.550.2MiniMax M2-herUnrankedMiniMax M355.7MiniMax M2.149.1MiniMax M2Unranked