Private AI

Nano GPT logo
NanoGPT
Nano GPT logo
NanoGPT
Nano GPT logo
NanoGPT
Search
 
 
 
 
All conversations

Create

Chat
Studio

Explore

Models
Pricing
AI Apps
Blog

Library

Gallery
Workspace
Usage & billing
API
Suggestions and bugs
Use cases

|

Terms

|

Privacy

|

Refunds
Usage & billing
Teams
Referrals
Presets
Settings
FAQPress & mediaBrand assetsBenchmarksEarn crypto
  1. Models
  2. Text
  3. MiniMax

MiniMax M3

MiniMax M3 is MiniMax's open-weights frontier model, built for coding, agent workflows, tool use, and multimodal understanding from step zero. Thinking stays off for faster direct answers unless a request asks for it; MiniMax M3 Thinking has it on by default. MiniMax reports 59.0% on SWE-Bench Pro and 66.0% on Terminal Bench 2.1, with Sparse Attention designed to scale context to 1M. It starts with a 512K context cap on NanoGPT for now.

  • Vision
  • Tool Calling
  • Structured Output

Added Jun 1, 2026

Model weights

Pricing

Auto routing · per 1M tokens
Input
$0.23
Output
$0.96
Cache read
$0.05
Compare provider prices

Specifications

Context window
512K
Max output
80K
Parameters
428B / 23B
Total / active
Avg output (7d)
820 tokens
Longer than 62% of models

Benchmarks

Sourced from LMArena.

Arena Score

1436.1

Overall Rank

#103 / 420

Votes

72,449

Confidence Interval

1432.4 - 1439.8

Category Scores

Coding

#89 / 415

20,253 votes

1493.9

Math

#118 / 404

3,578 votes

1425.4

Longer Query

#101 / 398

32,401 votes

1445.8

Creative Writing

#114 / 418

13,545 votes

1396.0

Instruction Following

#98 / 420

25,123 votes

1428.4

Hard Prompts

#107 / 420

47,827 votes

1457.7

Additional Categories
22

Industry Business And Management And Financial Operations

#88 / 413

13,847 votes

1445.2

Industry Software And It Services

#90 / 420

28,846 votes

1483.5

Multi Turn

#90 / 418

12,490 votes

1449.0

Hard Prompts English

#91 / 418

19,988 votes

1468.8

Industry Life And Physical And Social Science

#92 / 418

11,509 votes

1459.5

English

#94 / 420

29,313 votes

1450.6

Expert

#96 / 370

7,727 votes

1467.5

German

#97 / 320

1,226 votes

1437.1

Industry Mathematical

#99 / 401

4,052 votes

1444.9

Industry Writing And Literature And Language

#102 / 419

18,778 votes

1414.0

Exclude Ties

#103 / 420

54,761 votes

1432.0

French

#104 / 300

2,620 votes

1452.1

Polish

#104 / 244

1,270 votes

1427.9

Russian

#105 / 384

8,915 votes

1431.4

Industry Medicine And Healthcare

#108 / 390

4,816 votes

1452.1

Chinese

#109 / 394

4,733 votes

1470.5

Industry Entertainment And Sports And Media

#109 / 418

17,160 votes

1401.5

Non English

#110 / 420

43,125 votes

1420.0

Japanese

#113 / 288

1,089 votes

1377.7

Spanish

#116 / 303

2,276 votes

1422.6

Industry Legal And Government

#123 / 392

5,849 votes

1435.1

Korean

#123 / 288

1,201 votes

1366.3

Published 2026-11-08 · Matched as MiniMax-M3

LMArena Dataset

Providers

Choose or restrict providers with the API. Provider-selection prices include the selection markup. Provider selection guide · Pricing API (JSON, USD per 1K tokens)

Auto
Auto (ZDR)
$0.23$0.96$0.05N/AN/A
Auto
Auto (Anonymized)
$0.23$0.96$0.05N/AN/A
CoreWeave
CoreWeave
FP4
US
$0.242$1.01$0.05250.4s198 tps
DeepInfra
DeepInfra
FP8
US
$0.294$1.16$0.05882.1s27.5 tps
MiniMax
MiniMax
SG
$0.315$1.26$0.0631.5s78 tps
ModelRun
ModelRun
FP8
US
$0.315$1.36$0.05250.3s83 tps
Novita
Novita
FP8
US
$0.315$1.26$0.0631s116 tps
>524k$0.63$2.52$0.126
Parasail
Parasail
FP8
US
$0.315$1.26$0.0630.4s125 tps
StreamLake
StreamLake
FP8
CN
$0.315$1.26$0.0631.4s97 tps
Venice
Venice
FP8
US
$0.315$1.26$0.0630.8s75.5 tps
Ollama
Ollama
UNK
$0.63$2.52$0.126N/AN/A
SiliconFlow
SiliconFlow
FP8
SG
$0.63$2.52$0.126N/AN/A
Privacy:ZDR, prompts not retainedNo training on promptsRetention and training unclearBest latency and throughput are highlighted.
Batch API
ProviderInputOutputCached input
Auto
Autominimax/minimax-m3
$0.23/M$0.96/M$0.05/M

Batch rows for this model use normal routing and cost the same as standard API requests. Set provider in a row to choose a provider.

Related text models

Compare MiniMax M3 with similar models from the same provider or model family.

MiniMax M3.1 Flash

minimax/minimax-m3.1-flash

MiniMax M3.1 Flash is a multimodal model for coding, tool use, and long-context work. It accepts text, images, and video and produces text, with a 1,000,000-token context window and up to 524,288 output tokens within the remaining context. Thinking is always enabled, with low, medium, high, xhigh, and max effort levels; max is the default. Thinking tokens count toward the output limit. Temperature is fixed at 1 and top-p at 0.95; custom values have no effect.

MiniMax M3 Thinking

minimax/minimax-m3:thinking

MiniMax M3 Thinking is the adaptive-thinking version of MiniMax's open-weights frontier model for coding, agent workflows, tool use, long-context tasks, and native multimodal understanding. MiniMax reports 59.0% on SWE-Bench Pro and 66.0% on Terminal Bench 2.1, with Sparse Attention designed to scale context to 1M. It starts with a 512K context cap on NanoGPT for now.

MiniMax Latest

minimax/minimax-latest

Compatibility alias for MiniMax M3 Thinking, retaining its existing request format and structured-output support. The newer MiniMax M3.1 Flash is available separately as minimax/minimax-m3.1-flash and has different request parameter support.

MiniMax M2.7

minimax/minimax-m2.7

MiniMax M2.7 is the first model deeply involved in iterating on its own training. It excels in real-world software engineering (SWE-Pro 56.22%), end-to-end project delivery (VIBE-Pro 55.6%), and complex office workflows with strong tool-use compliance and agentic capabilities.

MiniMax M2.7 Turbo

minimax/minimax-m2.7-turbo

MiniMax M2.7 Turbo is the highspeed and higher priced route for M2.7.

MiniMax M2.5

minimax/minimax-m2.5

MiniMax M2.5 is a productivity-focused flagship model that builds on M2.1 with stronger coding and real-world office workflow performance (Word, Excel, PowerPoint), plus better tool-use planning and token efficiency.