Skip to main content

​
About the Provider

MiniMax is a Chinese AI research company focused on building large-scale open-source foundation models for coding, reasoning, and agentic workflows. Through its open-weights initiative, MiniMax develops efficient sparse models that deliver frontier-level performance accessible to developers and enterprises worldwide.

​
Model Quickstart

This section helps you quickly get started with the MiniMaxAI/MiniMax-M2.1 model on the Qubrid AI inferencing platform. To use this model, you need:
  • A valid Qubrid API key
  • Access to the Qubrid inference API
  • Basic knowledge of making API requests in your preferred language
Once authenticated with your API key, you can send inference requests to the MiniMaxAI/MiniMax-M2.1 model and receive responses based on your input prompts. Below are example placeholders showing how the model can be accessed using different programming environments.
You can choose the one that best fits your workflow. 
from openai import OpenAI

# Initialize the OpenAI client with Qubrid base URL
client = OpenAI(
    base_url="https://platform.qubrid.com/v1",
    api_key="QUBRID_API_KEY",
)

# Create a streaming chat completion
stream = client.chat.completions.create(
    model="MiniMaxAI/MiniMax-M2.1",
    messages=[
      {
        "role": "user",
        "content": "Explain quantum computing in simple terms"
      }
    ],
    max_tokens=8192,
    temperature=1,
    top_p=0.95,
    stream=True
)

# If stream = False comment this out
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)
print("\n")

# If stream = True comment this out
print(stream.choices[0].message.content)

​
Model Overview

MiniMax-M2.1 is a SOTA open-source coding and agentic model with 230B total parameters and only 10B active per token, achieving a 23:1 sparsity ratio.
  • Released December 2025, it achieves 74% on SWE-bench Verified — competitive with Claude Sonnet 4.5 — at a fraction of the cost, with best-in-class polyglot coding across Python, Java, Go, Rust, C++, TypeScript, and Kotlin.
  • With a 200K context window, FP8 native quantization, and open weights for local deployment, it is purpose-built for long-horizon agentic workflows and enterprise office automation.

​
Model at a Glance

FeatureDetails
Model IDMiniMaxAI/MiniMax-M2.1
ProviderMiniMax
ArchitectureSparse MoE Transformer — 230B total / 10B active per token (23:1 sparsity), FP8 quantization
Model Size230B Total / 10B Active
Parameters5
Context Length200K Tokens
Release DateDecember 2025
LicenseApache 2.0
Training DataLarge-scale multilingual code and instruction dataset across major programming languages

​
When to use?

You should consider using MiniMax-M2.1 if:
  • You need multilingual software development across Python, Java, Go, Rust, C++, TypeScript, and Kotlin
  • Your application requires long-horizon agentic coding workflows
  • You are building full-stack app generation pipelines
  • Your use case involves code review and optimization
  • You need office automation with complex multi-step tool use
  • You want Claude Sonnet 4.5-level coding performance at open-source cost

​
Inference Parameters

Parameter NameTypeDefaultDescription
StreamingbooleantrueEnable streaming responses for real-time output.
Temperaturenumber1Recommended at 1.0 for best performance.
Max Tokensnumber8192Maximum number of tokens the model can generate.
Top Pnumber0.95Controls nucleus sampling.
Top Knumber40Limits token sampling to top-k tokens.

​
Key Features

  • 74% SWE-bench Verified: Competitive with Claude Sonnet 4.5 on real-world software engineering tasks at open-source cost.
  • 23:1 Sparsity Ratio: Only 10B parameters active per token from 230B total — extreme efficiency for frontier-level coding performance.
  • Best-in-Class Polyglot Coding: Excels across Python, Java, Go, Rust, C++, TypeScript, and Kotlin for diverse software development workflows.
  • 200K Context Window: Supports long-horizon agentic tasks, full-codebase analysis, and extended multi-turn tool use.
  • FP8 Native Quantization: Reduced memory footprint with no accuracy trade-off for production deployments.
  • Open Weights: Fully available for local and on-premise deployment.

​
Summary

MiniMax-M2.1 is MiniMax’s flagship open-source coding and agentic model, delivering Claude Sonnet 4.5-level performance at open-source scale.
  • It uses a 230B sparse MoE Transformer with 10B active parameters per token and FP8 native quantization, released December 2025.
  • It achieves 74% on SWE-bench Verified with best-in-class polyglot coding across 7 major programming languages.
  • The model supports a 200K context window, long-horizon agentic workflows, and open weights for local deployment.
  • Licensed under Apache 2.0 for full commercial use.