MiniMax M2.1 - MiniMax's multi-language programmable AI model

What is MiniMax M2.1?

MiniMax M2.1 is a multi-language programming AI model launched by MiniMax. The model systematically enhances capabilities in multiple languages, including Rust, Java, and Golang, covering development from the lowest level to the application layer. In web and app development, the model significantly improves design understanding and aesthetic expression, optimizes the execution of compound instructions, and makes it more practical for office scenarios. Compared to its predecessor, MiniMax M2, M2.1 offers more concise and efficient responses and performs stably and excellently across various programming tools and frameworks. The model possesses high-quality dialogue and writing capabilities, helping businesses and individuals achieve a more AI-native work and lifestyle.

Main functions of MiniMax M2.1

• Multilingual programming supportMiniMax M2.1 enhances the capabilities of multiple languages such as Rust, Java, and Golang, covering development from the underlying system to the application layer, and performs exceptionally well in web and mobile development.
• Complex task processingBy introducing Interleaved Thinking, M2.1 can efficiently integrate compound instructions, support building complete applications from scratch, and optimize code generation and performance.
• Office automationThe M2.1 can control the mouse and keyboard through natural language commands to complete complex office tasks such as data processing and document generation, achieving end-to-end automation.
• Efficient interaction and optimizationThe model offers concise and efficient responses, faster response times, reduced token consumption, and automatic caching support, thus optimizing the user experience.
• Generalization capabilities of tools and frameworksIt performs stably across various programming tools and agent frameworks, supports multiple context management mechanisms, and provides APIs of different speeds to meet diverse needs.
• High-quality dialogue and writingIn everyday conversations and technical writing, M2.1 provides more detailed and structured answers, not just limited to code generation.

MiniMax M2.1 performance

• Software Engineering Core Rankings PerformanceThe MiniMax M2.1 shows a significant improvement over the M2 in core benchmarks for software engineering scenarios, especially in multilingual scenarios, surpassing Claude Sonnet 4.5 and Gemini 3 Pro, and approaching the level of Claude Opus 4.5.
• Framework generalization and stabilityIn the SWE-bench Verified test, MiniMax M2.1 demonstrated good framework generalization and stability, and is suitable for various coding agent frameworks.
• Detailed scene performanceIn specific scenarios such as test case generation, code performance optimization, code review (SWE-Review), and instruction compliance (OctoCodingBench), MiniMax M2.1 shows comprehensive improvement over M2, matching or exceeding Claude Sonnet 4.5.
• VIBE Benchmark PerformanceIn the VIBE benchmark covering Web, simulation, Android, iOS, and backend, the MiniMax M2.1 achieved an average score of 88.6, close to Claude Opus 4.5, and significantly outperformed Claude Sonnet 4.5 on almost all subsets.
• Office environment and comprehensive capabilitiesMiniMax M2.1 has steadily improved in office scenarios, long-term tool calls, and comprehensive intelligence index, demonstrating the model's ability to solve more complex real-world tasks.

MiniMax M2.1 project address

• Project official websiteMiniMax
• HuggingFace model libraryhttps://huggingface.co/MiniMaxAI/MiniMax-M2.1

How to use MiniMax M2.1

• API callsUsers can directly access the model through the MiniMax M2.1 API open platform: https://platform.minimaxi.com/docs/guides/text-generation
• MiniMax Agent useGet started quickly with the general agent product (MiniMax Agent).
• Local deploymentThe open-source version supports local deployment (Hugging Face model library https://huggingface.co/MiniMaxAI/MiniMax-M2.1), and the weights will be fully open-sourced in a few days.

To meet different needs, MiniMax offers two API versions: M2.1 and M2.1-lightning. The latter is faster and suitable for high TPS scenarios. MiniMax M2.1 also fully supports automatic caching to optimize the user experience.

MiniMax M2.1 product pricing

• MiniMax – M2.1:
  • Enter price2.1 yuan/million tokens
  • Output Price8.4 yuan/million tokens
  • Cache read0.21 yuan/million tokens
  • Cache write2.625 yuan/million tokens
• MiniMax – M2.1 – lightning(Faster version):
  • Enter price2.1 yuan/million tokens
  • Output Price16.8 yuan/million tokens
  • Cache read0.21 yuan/million tokens
  • Cache write2.625 yuan/million tokens

Application scenarios of MiniMax M2.1

• Multilingual Programming and DevelopmentThe model supports programming languages such as Rust, Java, and Golang, and is suitable for web, mobile application (Android/iOS) and backend development, improving development efficiency.
• Office AutomationThe model can automatically process data, generate documents, and schedule tasks. It can also control the mouse and keyboard through natural language commands, thus optimizing office workflows.
• Complex task processingIt efficiently integrates multi-step instructions, processes complex logical tasks, and ensures task continuity and accuracy.
• Code generation and optimizationThe model can generate high-quality code, optimize the performance of existing code, and assist in development and testing.
• Full-stack application developmentSupports building web, mobile, and backend applications from scratch, covering the full-stack development process.