Chinese artificial intelligence startup MiniMax today announced the release of M2.1, which significantly enhances the performance of complex real-world tasks and agent functions across more programming languages and office scenarios.
Key highlights of M2.1 include significantly enhanced programming skills in a number of programming languages, including Rust, Java, Golang, C++, Kotlin, Objective-C, TypeScript, and JavaScript. Significantly improve your aesthetic design abilities and understanding of user interface development for Web, Android, and iOS.
M2.1 upgrades systematic problem solving over M2 and focuses not only on the correct execution of code, but also on following instructions with additional or complex guidance. The company says this provides greater ease of use in real-world office situations, where even simple rules can have complex nuances.
To achieve this, the company focused not only on strong coding abilities, but also on strengthening dialogue and writing skills. This model excels in everyday conversation, technical documentation, and writing, along with the ability to provide structured responses.
“Our users have come to rely on MiniMax for frontier-grade coding assistance at a fraction of the cost,” said Scott Breitenother, co-founder and CEO of Kilo Code Inc., an open source agent-based AI coding agency. “Early testing shows M2.1 excels in everything from architecture and orchestration to code review and deployment.”
MiniMax M2 was released in late October this year. The company says M2.1 has proven to have significant improvements in functionality compared to its predecessor, especially in multilingual scenarios. Here it beats Anthropic PBC's Claude Sonnet 4.5 and approaches Claude Opus 4.5 (a larger, more complex model).
As part of the evaluation, MiniMax established a new benchmark, VIBE (Visual and Interactive Benchmark for Execution). The suite included five core capabilities: web, simulation, Android, iOS, and backend development. Unlike other benchmarks, VIBE consists of an agent language as the verifier. This allows you to evaluate the interactive logic and visual aesthetics of the generated application.
The M2.1 showed what the company calls “outstanding performance” in the VIBE benchmark, achieving an average score of 88.6. It particularly excelled in the VIBE-Web and VIBE-Android subsets, scoring 91.5 and 89.7, respectively.
The company also tested the new model across comprehensive industry benchmarks for both coding and knowledge, including MMLU-Pro, Humanity's Last Exam, and Toolathon (for AI agents), against leading vendors such as Anthropic, Google LLC, OpenAI Group PBC, and DeepSeek.
The model showed consistently high performance in agent tool use, real-world knowledge, and complex problem-solving ability. I got a score of 22.0 on HLE without tools. This is a challenging academic benchmark that includes thousands of graduate-level, multimodal questions across a diverse range of topics. On MMLU, a similarly comprehensive subject knowledge benchmark, this model scored 88, consistently matching or close to the flagship Frontier model.
This model is available as an application programming interface from MiniMax or as an open weight download from HuggingFace. This page is currently not yet available. The company's flagship service, MiniMax Agent, is built on the new MiniMax-2.1.
Images: Canva, MiniMax
Support our mission of keeping content open and free by joining the theCUBE community. Join theCUBE’s Alumni Trust Networka place where technology leaders connect, share intelligence, and create opportunities.
- over 15 million viewers of theCUBE videospowering conversations across AI, cloud, cybersecurity, and more
- 11.4k+ theCUBE Alumni — Connect with over 11,400 technology and business leaders who are shaping the future through our trusted, unique network.
About SiliconANGLE Media
Founded by technology visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach more than 15 million elite technology professionals. Our new, proprietary theCUBE AI Video Cloud leverages theCUBEai.com neural networks to deliver breakthrough advances in audience interaction, helping technology companies make data-driven decisions and stay at the forefront of industry conversations.
