Chinese AI startup MiniMax has released M2.1, an upgraded model that brings significant performance gains for real-world tasks and expanded programming capabilities across multiple languages.
The release comes as MiniMax prepares for its Hong Kong IPO in January 2026. The company cleared its stock exchange listing hearing on December 17 and aims to raise up to $637 million at a $4 billion valuation. Backed by Alibaba, Tencent, and Hillhouse Investment, MiniMax has positioned itself among China’s top AI contenders alongside DeepSeek and Moonshot AI.
Enhanced Coding and Design Capabilities
M2.1 delivers dramatically improved programming skills across Rust, Java, Golang, C++, Kotlin, Objective-C, TypeScript, and JavaScript. The model also gains stronger UI/UX comprehension for web, Android, and iOS development.
A key upgrade focuses on systematic problem-solving. M2.1 handles not just code execution correctness but also complex instructions with nuanced guidance. This makes it more practical for office scenarios where simple rules often carry hidden complexity.
New VIBE Benchmark Introduced
MiniMax established a new evaluation standard called VIBE (Visual and Interactive Benchmark for Execution). The suite tests five core capabilities: web, simulation, Android, iOS, and backend development.
Unlike traditional benchmarks, VIBE uses an agent-as-a-verifier approach to assess both interactive logic and visual aesthetics of generated applications.
M2.1 achieved an average VIBE score of 88.6, with particularly strong results in VIBE-Web (91.5) and VIBE-Android (89.7).
Competitive Positioning
The company tested M2.1 against models from Anthropic, Google, OpenAI, and DeepSeek across industry benchmarks including MMLU-Pro, Humanity’s Last Exam, and Toolathon for AI agents.
MiniMax’s previous M2 model reached the global top five on Artificial Analysis rankings in October, placing ahead of Google DeepMind’s Gemini 2.5 Pro among open-source models. M2.1 builds on that foundation with targeted improvements for agentic workflows and multi-language coding tasks.
The Shanghai-based company reports 27.6 million monthly active users as of September 2025 and over 200 million cumulative users across more than 200 countries.
