MiniMax M3

Tool callingStructured outputReasoningImage inputVideo input

Hugging Face Website

MiniMax multimodal model for long-context coding, perception, and agent planning

Context: 1.0M
Released: 4mo ago
Parameters: 0
Input / 1M tokens: $0.30
Output / 1M tokens: $1.20
Cached input / 1M tokens: $0.06

Sources:models.devOpenRouter

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

Other published benchmarks

Evaluation	Reported score and source
GPQA diamond	90.9% Epoch AI · MiniMax-M3 Epoch AI evaluation
GPQA diamond	81.3% Epoch AI · MiniMax-M3_none Reasoning effort: none Epoch AI evaluation
OSWorld 2.0	4.6% https://osworld-v2.xlang.ai/ · MiniMax-M3 https://osworld-v2.xlang.ai/ Original-source terms apply
OTIS Mock AIME 2024-2025	71.1% Epoch AI · MiniMax-M3 Epoch AI evaluation
OTIS Mock AIME 2024-2025	26.7% Epoch AI · MiniMax-M3_none Reasoning effort: none Epoch AI evaluation

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Specifications

Developer: MiniMax
Model family: MiniMax
Type: Language model
Predecessor: —
Successor: —
Released: June 1, 2026
Model ID: minimax/minimax-m3
Openness: Open weights
License: —
Parameters: 0
Architecture: minimax-m3
Families: minimax-m3
Input modalities: text, image, video
Output modalities: text
Modality: text+image+video->text
Native context: 512K
Tokenizer: Other
Max output tokens: 512K
Hugging Face: MiniMaxAI/Minimax-M3
Source freshness: Metadata synced Oct 8, 2026, 10:17 PM UTC

Open-weight builds

Ollama variants

Every published Ollama tag matched to this model, including parameter editions, instruction/base variants, and quantized builds.

1 builds

Variant	Parameters	Type	Quantization	Size	Context	Input
minimax-m3:cloud	0	cloud	—	—	512K	text + image

Additional source evidence

Source model: minimax/MiniMax-M3
Source family: minimax
Last updated: 2026-06-01
Model context: 1.0M
Model output limit: 512K
Open weights: Yes
File attachments: Yes
Reasoning: Yes
Tool calling: Yes
Temperature control: Yes

Downloadable weights

Hugging Face

Published benchmarks

Benchmark	Score	Evaluation details	Evidence
SWE-Bench Verified	80.5 · resolved	Claude Code · 2026-06-01	Source
SWE-Bench Pro	59 · resolve rate	Claude Code · 2026-06-01	Source
Terminal-Bench	66 · success rate	2.1 · 2026-06-01	Source
BrowseComp	83.52 · accuracy	2026-06-01	Source
MCP Atlas	74.2 · success rate	2026-06-01	Source
OSWorld-Verified	70.06 · success rate	2026-06-01	Source

Complete source record (JSON)

{
  "id": "minimax/MiniMax-M3",
  "name": "MiniMax-M3",
  "description": "MiniMax multimodal model for long-context coding, perception, and agent planning",
  "family": "minimax",
  "last_updated": "2026-06-01",
  "attachment": true,
  "reasoning": true,
  "tool_call": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text",
      "image",
      "video"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 1048576,
    "output": 512000
  },
  "benchmarks": [
    {
      "name": "SWE-Bench Verified",
      "score": 80.5,
      "source": "https://www.minimax.io/blog/minimax-m3",
      "metric": "resolved",
      "harness": "Claude Code",
      "date": "2026-06-01"
    },
    {
      "name": "SWE-Bench Pro",
      "score": 59,
      "source": "https://www.minimax.io/blog/minimax-m3",
      "metric": "resolve rate",
      "harness": "Claude Code",
      "date": "2026-06-01"
    },
    {
      "name": "Terminal-Bench",
      "score": 66,
      "source": "https://www.minimax.io/blog/minimax-m3",
      "metric": "success rate",
      "version": "2.1",
      "date": "2026-06-01"
    },
    {
      "name": "BrowseComp",
      "score": 83.52,
      "source": "https://www.minimax.io/blog/minimax-m3",
      "metric": "accuracy",
      "date": "2026-06-01"
    },
    {
      "name": "MCP Atlas",
      "score": 74.2,
      "source": "https://www.minimax.io/blog/minimax-m3",
      "metric": "success rate",
      "date": "2026-06-01"
    },
    {
      "name": "OSWorld-Verified",
      "score": 70.06,
      "source": "https://www.minimax.io/blog/minimax-m3",
      "metric": "success rate",
      "date": "2026-06-01"
    }
  ],
  "release_date": "2026-06-01",
  "open_weights": true,
  "weights": [
    {
      "label": "Hugging Face",
      "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3"
    }
  ]
}

Infrastructure

Provider availability

All providers

Provider	Endpoint	Context	Input /M tokens	Output /M tokens	Cached input /M tokens	Cache write /M tokens	Uptime 1d	Data handling
CoreWeave	coreweave/fp4	262K	$0.23	$0.96	$0.05	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
GMICloud	gmicloud/fp8	1.0M	$0.24	$0.96	$0.05	—	99.72%	ZDR: No Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
DeepInfra	deepinfra/fp8	524K	$0.28	$1.10	$0.06	—	99.91%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
AtlasCloud	atlas-cloud/fp8	524K	$0.30	$1.20	$0.06	—	99.45%	ZDR: No Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Minimax	minimax/fp8	524K	$0.30	$1.20	$0.06	—	99.93%	ZDR: No Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
ModelRun	modelrun/fp8	1.0M	$0.30	$1.30	$0.05	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Novita	novita/fp8	1M	$0.30	$1.20	$0.06	—	99.91%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Parasail	parasail/fp8	1.0M	$0.30	$1.20	$0.06	—	99.57%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
StreamLake	streamlake/fp8	1M	$0.30	$1.20	$0.06	—	99.58%	ZDR: No Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Together	together	524K	$0.30	$1.20	$0.06	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Venice	venice/fp8	524K	$0.30	$1.20	$0.06	—	99.35%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Mara	mara	1.0M	$0.60	$2.40	—	—	98.28%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
SambaNova	sambanova	1.0M	$0.60	$2.40	$0.06	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15

Provider data:OpenRouter

Deployment graph

Can I run this?

Vertical Tabs only shows compatibility when it has structured evidence tying this exact model to hardware, runtime, quantization, memory use, and observed performance.

Browse hardware