MiniMax M3.1 Flash

Native multimodal model for coding, agentic workflows, and computer use

Try now

Read Docs

About model

MiniMax M3.1 Flash is a native multimodal Mixture-of-Experts model from MiniMax for coding, agentic workflows, computer use, and cybersecurity. With roughly 428B total parameters and 23B active per token, it accepts text, image, and video input across a context window of up to 1M tokens. Building on MiniMax M3, it improves both task performance and inference efficiency: MiniMax Sparse Attention now runs throughout the model, quantization-aware training covers the routed experts and attention paths to reduce memory footprint, and a DSpark speculative decoding head ships with the release. The model carries complex, multi-step professional workflows through to completion, understands real codebases, fixes bugs, and builds features, and reasons over images and charts to produce polished frontends and presentations. Released under the MiniMax Community License. Available on Together AI.

‍

Total Parameters (23B active)

428B

Native multimodal MoE built for agentic and engineering work

Context Window

Long agent sessions, large codebases, and long documents in one request

Text, Image & Video Input

Native

Reasons over screens, charts, and video alongside text

Model key capabilities

Cowork Automation: Built for complex, multi-step workflows across professional tasks, from everyday questions to financial analysis
Real-Codebase Engineering: Understands real codebases, fixes bugs, builds and refactors features, and generates full repositories
Computer Use & Visual Work: Operates computer environments from visual input, reasons over images and charts, and produces polished frontends and presentations
Production-Ready Infrastructure: 99.9% SLA, available on serverless and dedicated infrastructure

Performance benchmarks

Model	FrontierMath Tier 4	GPQA Diamond	HLE	SciCode	GDPval-AA	Terminal-Bench 2.1	Agent Arena	FrontierCode	DeepSWE
MiniMax M3.1 Flash			43.4						73.7
Related open-source models
Competitor closed-source models
Claude Fable 5	87.8%	92.6%	53%	60%	62%	85%		53.5%	70%
Claude Opus 5	73.2%	93.2%	53%	56%	68%	89%	+16.4pp	53.4%	74%
GPT-5.6 Sol	82.9%	94.1%	47%	56%	61%	88%		47.5%	73%
Grok 4.5	24.4%	93.1%	40%	54%	51%	82%	+4.2pp	42.4%	54%
GPT-5.6 Luna	61.0%	91.1%	37%	53%	54%	81%		39.8%	67%

API usage

cURL
Python
Typescript

Endpoint:

MiniMaxAI/MiniMax-M3.1-Flash

curl -X POST "https://api.together.xyz/v1/chat/completions" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMaxAI/MiniMax-M3.1-Flash",
    "messages": [
      {
        "role": "user",
        "content": "What are some fun things to do in New York?"
      }
    ]
}'

from together import Together

client = Together()

response = client.chat.completions.create(
  model="MiniMaxAI/MiniMax-M3.1-Flash",
  messages=[
    {
      "role": "user",
      "content": "What are some fun things to do in New York?"
    }
  ]
)
print(response.choices[0].message.content)

import Together from 'together-ai';
const together = new Together();

const completion = await together.chat.completions.create({
  model: 'MiniMaxAI/MiniMax-M3.1-Flash',
  messages: [
    {
      role: 'user',
      content: 'What are some fun things to do in New York?'
     }
  ],
});

console.log(completion.choices[0].message.content);

Model card
Architecture Overview:
• Native multimodal Mixture-of-Experts with approximately 428B total parameters and 23B activated per token, retaining MiniMax M3's core MoE architecture
• MiniMax Sparse Attention (MSA) throughout the model, with the three full-attention layers from M3 replaced by sparse attention
• DSpark speculative decoding draft head included, replacing M3's earlier multi-token prediction approach
• Text, image, and video input with text output, across a context window of up to 1M tokens

Training Methodology:
• Builds on MiniMax M3, improving task performance and inference efficiency
• Quantization-aware training applied to the routed MoE experts and attention paths, reducing memory footprint and improving serving efficiency

Performance Characteristics:
• Workflow automation (AutomationBench v1.0.6: 62.8, the highest among the open models in MiniMax's comparison; BankerToolBench financial tool use: 76.2; GDP.pdf AA: 20.8)
• Agentic tasks (Agents' Last Exam: 29.6; LOCA-Bench at 256K context: 60.3; PostTrain Bench v1.1: 39.3)
• Computer use (OSWorld 2.0: 61.7)
• Software engineering (DeepSWE v1.1: 73.7; NL2Repo-Bench repository generation: 61.2; Terminal Bench 4.0: 38.1; SWE Refactor Bench: 23.0)
• Security research (CyberGym vulnerability discovery: 82.2)
• Visual reasoning (BabyVision with tools: 91.0; CharXiv RQ scientific chart reasoning: 82.5)
• Knowledge and reasoning (Humanity's Last Exam without tools: 43.4; MiniMax-reported AA-Omniscience: 56% and CritPt research-level physics: 22.3%)
‍
Prompting
Together AI API Access:
• Access MiniMax M3.1 Flash via Together AI APIs using the endpoint MiniMaxAI/MiniMax-M3.1-Flash
• Authenticate using your Together AI API key in request headers
• Set reasoning_effort to low, medium, high, xhigh, or max to control deliberation depth
• Recommended sampling: temperature 1.0 and top-p 0.95
• Supports text, image, and video input across a context window of up to 1M tokens
• Available on Together AI serverless and dedicated infrastructure
‍
Applications & use cases
Cowork & Workflow Automation:
• Automate complex, multi-step professional workflows end to end
• Run financial analysis tasks for banking and finance teams through tool use
• Sustain long agent sessions with up to 1M tokens of context

Software Engineering:
• Navigate real codebases to fix bugs, refactor, and build features
• Generate full repositories from natural language specifications
• Produce polished frontends from requirements or visual references

Computer Use & Security:
• Automate desktop and application workflows grounded in screen understanding
• Run authorized vulnerability discovery across codebases
• Analyze charts, documents, and video inside agent pipelines
‍