Introducing the AI model 'MiniMax M2.1,' which outperforms Gemini 3.0 Pro and GPT-5.2 in some tests.

AI company MiniMax has announced the release of the multimodal model ' MiniMax M2.1 .' It claims to be 'the most powerful open source model in the agent era,' and has even outperformed Gemini 3 Pro and Claude 4.5 Sonnet in some benchmarks.

MiniMax M2.1: Significantly Enhanced Multi-Language Programming, Built for Real-World Complex Tasks - MiniMax News

MiniMax M2.1 is officially live🚀
Built for real-world coding and AI-native organizations — from vibe builds to serious workflows.

A SOTA 10B-activated OSS coding & agent model, scoring 72.5% on SWE-multilingual and 88.6% on our newly open-sourced VIBE-bench, exceeding leading… pic.twitter.com/pnQG53H7rM

— MiniMax (official) (@MiniMax__AI) December 23, 2025

The features of MiniMax M2.1 are as follows:

◆Excellent support for multiple programming languages
While many previous AI models have focused on optimizing for Python, in reality, systems are generally built using multiple programming languages, not just Python. MiniMax M2.1 systematically strengthens its support for multiple programming languages, including Rust, Java, Golang, C++, Kotlin, Objective-C, TypeScript, and JavaScript, resulting in industry-leading overall performance in multilingual tasks and complete coverage from low-level systems to application development.

◆ Strengthening web and app development
It significantly strengthens native Android and iOS development capabilities, addressing the weaknesses of mobile development, while systematically strengthening design comprehension and aesthetic expression capabilities, enabling simulation and high-quality visualization of 3D scientific scenes. It also evolves vibecoding into a sustainable and practical production method.

◆Realizing office scenarios through strengthened 'composite command constraints'
As one of the first open source model series to systematically implement 'interleaved thinking,' which was introduced in Claude, MiniMax M2.1 further strengthens its systematic problem-solving capabilities. By focusing not only on the accuracy of code execution but also on the integrated execution of 'composite command constraints,' it offers greater practicality in realistic office scenarios.

◆ Responses are concise and efficient
Compared to the MiniMax M2, the MiniMax M2.1 has a simpler response and thought chain, significantly improving response speed and drastically reducing token consumption in programming and dialogue, resulting in smoother and more efficient performance in AI coding and agent-driven continuous workflows.

◆Excellent generalization ability of agent tools
MiniMax M2.1 provides consistency and stability across various programming tools and agent frameworks, including Claude Code, Droid (Factory AI), Cline, Kilo Code, Roo Code, and BlackBox, and provides reliable support for context management mechanisms such as Skill.md.

◆High-quality dialogue and text generation
It goes beyond simply being 'good at coding' to providing detailed and structured responses in everyday conversations, writing documents, and other tasks.

Below is a graph comparing the MiniMax M2.1 with the M2, DeepSeek V3.2, Claude Sonnet 4.5, Gemini 3 Pro, and GPT-5.2 (thinking) in 10 benchmark tests. In multiple tests, the MiniMax M2.1 recorded higher scores than the others.

SOTA across SWE-Verified, SWE-Multilingual, Multi-SWE, VIBE-Bench, and Terminal-Bench 2.0. pic.twitter.com/tUj4qzzb9D

— MiniMax (official) (@MiniMax__AI) December 23, 2025