MiniMax M3

Name: MiniMax M3
Author: MiniMax

Multimodal

MiniMax

MiniMax M3 is a 428B-parameter multimodal MoE model released in June 2026. It combines strong document understanding with long-context support of up to 1M tokens. Weights are available on Hugging Face under the MIT license.

Key Specifications

Parameters

428.0B

Context

1.0M

Release Date

June 1, 2026

Average Score

58.4%

API Documentation Research Paper Results Blog

Timeline

Key dates in the model's history

Announcement

June 1, 2026

Last Update

August 27, 2026

Today

October 22, 2026

Technical Specifications

Parameters

428.0B

Training Tokens

Knowledge Cutoff

Family

Capabilities

MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)

$0.30

Output (per 1M tokens)

$1.20

Max Input Tokens

1.0M

Max Output Tokens

131.1K

Supported Features

Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Programming

Programming skills tests

SWE-Bench Verified

• Self-reported

80.5%

Other Tests

Specialized benchmarks

OmniDocBench 1.5

Self-reported by the model provider • Self-reported

92.0%

APEX-Agents

• Self-reported

27.7%

BankerToolBench

• Self-reported

76.1%

BrowseComp

• Self-reported

83.5%

CL-bench

• Self-reported

20.5%

Claw-Eval

Pass^3 • Self-reported

74.5%

DRACO

• Self-reported

73.2%

Finance Agent v2

• Verified

48.3%

FrontierCode 1.1

FrontierCode 1.1 current leaderboard; mergeability score: 14.7%. • Verified

14.7%

GDPval-Rubrics

• Self-reported

74.8%

IMO 2025

• Self-reported

83.3%

KernelBench Hard

• Self-reported

28.8%

LiveBench

2026-01-08 • Verified

70.0%

LiveSQLBench

• Self-reported

40.2%

LOCA-Bench (256k)

• Self-reported

49.3%

MCP Atlas

• Self-reported

74.2%

MMMU-Pro

• Self-reported

78.1%

NL2Repo

• Self-reported

42.1%

OfficeQA Pro

• Self-reported

45.1%

OSWorld-Verified

Max Steps = 200 • Self-reported

70.1%

PaperBench

• Self-reported

52.6%

PostTrainBench

• Self-reported

37.1%

SpreadSheetBench-v1

• Self-reported

89.3%

SVG-Bench

• Self-reported

63.7%

SWE Atlas - Codebase QnA

• Self-reported

37.9%

SWE Atlas - Test Writing

• Self-reported

30.8%

SWE-Bench Pro

• Self-reported

59.0%

SWE-fficiency

• Self-reported

34.8%

Terminal-Bench 2.1

• Self-reported

66.0%

USAMO 2026

• Self-reported

85.7%

VIBE-V2

• Self-reported

50.1%

Video-MME

With subtitles • Self-reported

85.4%

VideoMMMU

• Self-reported

84.6%

YC-Bench

Final assets (fund), USD • Self-reported

21.0%

License & Metadata

License

mit

Announcement Date

June 1, 2026

Last Updated

August 27, 2026

Similar Models

All Models

MiniMax M2.5

MiniMax

MM230.0B

Released:Feb 2026

MiniMax M2

MiniMax

230.0B

Best score:0.8 (GPQA)

Released:Oct 2025

Price:$0.30/1M tokens

Mistral Large 4

Mistral AI

MM1.1T

Released:Oct 2026

Price:$0.68/1M tokens

MiMo-V2.6-Pro

Xiaomi

MM1.0T

Released:Sep 2026

Price:$0.43/1M tokens

Kimi K2.7 Code

Moonshot AI

MM1.0T

Released:Jun 2026

Price:$0.95/1M tokens

Step 3.7 Flash

StepFun

MM198.0B

Released:Jun 2026

Price:$0.20/1M tokens

MAI-Code-1.1-Flash

Microsoft

MM138.0B

Released:Aug 2026

Price:$0.20/1M tokens

Mistral Medium 3.5

Mistral AI

MM128.0B

Released:Apr 2026

Price:$1.50/1M tokens

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.