Model

MiniMax M3

OPEN

MiniMax · MiniMax M3 family · released Jun 1, 2026

Successor to M2 introducing MiniMax Sparse Attention (MSA), giving a native 1M-token multimodal (text/image/video input) context window with reported 9x faster prefill / 15x faster decode vs M2 at 1M context. Open-weight release on Hugging Face alongside hosted API access. Official pricing page lists a base rate of $0.60/$2.40 per million input/output tokens with a standing 50% promotional discount to the effective $0.30/$1.20 used here for inputs ≤512K tokens; price doubles beyond 512K tokens and a 1.5x 'priority' service tier also exists (not modeled as a separate entry).

ReasoningCodingVisionFunction callingTool useAgentic

2519.3

Elo · rank #14

Parameters

428B

Active params

23B (MoE)

Context

1.024M tokens

Architecture

Sparse Mixture-of-Experts with MiniMax Sparse Attention (MSA) for 1M-token context, native multimodal (mixed-modality pretraining from step one); 428B total / 23B active params

License

MiniMax Model Community License (non-commercial restrictions apply; HF license field shows 'minimax-community') — verify exact license text at https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE

Languages

—

API price (in/out)

$0.3 / $1.2

Modalities

text · vision · video

Benchmark results

Bar shows position within the tracked field; marker = field best

BrowseCompAgents83.5%#11

best: Kimi K3 · 91.2%

GPQA DiamondReasoning92.9%#11

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning37.0%#31

best: Claude Sonnet 5 · 57.4%

IFBenchReasoning83.0%#1

best: this model · 83.0%

MMMU-ProVision78.1%#16

best: Claude Opus 4.7 · 85.5%

OSWorld-VerifiedAgents75.2%#8

best: Claude Fable 5 · 85.0%

SWE-bench ProCoding59.0%#10

best: Claude Fable 5 · 80.0%

SWE-bench VerifiedCoding80.5%#15

best: Claude Fable 5 · 95.0%

Terminal-Bench 2.0Coding66.0%#23

best: GPT-5.6 · 88.8%

USAMOMath85.7%#2

best: Claude Opus 4.8 · 96.7%

Video-MMEVision85.4%#8

best: Seed 2.1 Pro · 89.2%

Run it locally

VRAM @ Q4

—

VRAM @ FP16

—

Fits on (Q4)

Multi-node cluster required

Throughput data unavailable.

Quantizations

—

MiniMax M3 family

Elo progression across releases

MiniMax M3Jun 20262519.3

Resources

HFMiniMaxAI/MiniMax-M3 DOCShttps://platform.minimax.io/docs/release-notes/models APIAvailable via MiniMax API

API price $0.3/$1.2 · each benchmark row carries its own source badge (see methodology)