In a strategic move to capture attention ahead of their upcoming initial public offerings (IPOs) in Hong Kong, two leading Chinese start-ups, Zhipu AI and MiniMax, have launched their latest artificial intelligence models almost simultaneously. Both companies, referred to as China’s “AI tigers,” are vying for dominance in the open-source AI landscape. The release of these advanced AI models not only highlights the companies’ technological progress but also signals their competitive edge in the rapidly growing global AI market.
Zhipu AI: The Launch of GLM-4.7
Zhipu AI, a Beijing-based AI start-up, known internationally as Z.ai, unveiled its new GLM-4.7 model on Monday evening. The model features enhanced coding abilities and is touted as narrowing the performance gap with major U.S. competitors like Google DeepMind and Anthropic. Zhipu AI claims that GLM-4.7 has achieved high scores in various coding benchmarks, even surpassing OpenAI’s GPT-5, which was released earlier in August 2025.
According to Zhipu AI, the GLM-4.7 model excels in software engineering tasks, surpassing the performance of previous models in real-world benchmarks. Notably, GLM-4.7 outperformed OpenAI’s GPT-5 in tests conducted using SWE-Bench, a benchmark that evaluates AI models on real-world software engineering problems. This puts Zhipu AI in direct competition with the likes of Anthropic’s Claude Opus 4.5 and Google’s Gemini Pro 3, both of which had previously set high performance standards.
One of the major breakthroughs for Zhipu AI is its ability to perform exceptionally well on SWE-Bench, a benchmark that evaluates AI models on coding tasks faced by developers in the software industry. Zhipu AI proudly announced that GLM-4.7 had crossed the 70% threshold on SWE-Bench, a feat previously achieved only by Anthropic’s Claude Opus 4 model. This development is significant as it showcases Zhipu AI’s improved capabilities in coding agents—a domain previously dominated by large U.S. companies.
Despite these technical benchmarks, Zhipu AI emphasized that benchmarks alone don’t define a model’s performance. In a blog post, the company encouraged developers and users to explore how the GLM-4.7 model performs in real-world scenarios, highlighting its ease of use and versatility in software development.
MiniMax: The M2.1 Model
Just hours after Zhipu AI’s announcement, MiniMax, a Shanghai-based AI start-up, introduced its M2.1 model. The M2.1 also boasts impressive improvements in coding abilities, particularly when tested internally. Like Zhipu’s GLM-4.7, MiniMax’s M2.1 model still lags behind Claude Opus 4.5 but offers a unique set of capabilities.
One of the standout features of the M2.1 model is its industry-leading performance in processing multiple programming languages simultaneously. MiniMax emphasized that its model could efficiently handle languages like Java, Rust, and Golang—key programming languages for modern software development. This flexibility positions MiniMax as a competitive player, especially in industries where the ability to process multiple languages quickly is crucial for productivity and development speed.
Additionally, MiniMax introduced a new in-house benchmark, called VIBE, which assesses AI models based on their ability to “vibe code.” This process involves using generative AI to help amateur programmers create custom software applications through simple language prompts. MiniMax’s M2.1 model scored highly on the VIBE benchmark, further solidifying its reputation in the AI coding space.
Open-Source Approach and Weights Release
In terms of accessibility, Zhipu AI has already released the weights of its GLM-4.7 model, making it available for developers around the world to deploy locally on their own hardware. This open-source approach allows for wider adoption and further enhances the company’s appeal, especially to developers who seek customizable AI solutions.
MiniMax, on the other hand, has announced plans to release the weights for its M2.1 model later this week. These weights are crucial as they represent the variables that encode the model’s “intelligence” and are adjusted during training. The release of these weights is expected to enable developers to experiment with the M2.1 model, further expanding its potential applications in coding and software development.
Strategic Implications for Zhipu AI and MiniMax
The simultaneous release of these AI models by Zhipu AI and MiniMax underscores the companies’ ambitions to become global leaders in the open-source AI market. With their IPOs in Hong Kong on the horizon, both start-ups are keen to attract investors by showcasing their technological capabilities and commitment to advancing AI innovation.
As the AI sector continues to grow rapidly, the competition between these Chinese start-ups and global giants like Google, OpenAI, and Anthropic is heating up. The near-simultaneous release of the GLM-4.7 and M2.1 models serves as a reminder of China’s growing presence in the AI field. With advancements in coding abilities, new benchmarks, and open-source accessibility, both Zhipu AI and MiniMax are well-positioned to challenge established players and attract attention from global investors.
The release of Zhipu AI’s GLM-4.7 and MiniMax’s M2.1 models highlights the rapid innovation and competition in the artificial intelligence space. As China’s “AI tigers” prepare for their IPOs, they have demonstrated their commitment to becoming leaders in the global AI market by releasing models with impressive coding capabilities and technological advancements. As the world watches, these companies are poised to play a pivotal role in shaping the future of AI development, while contributing to China’s growing influence in the global tech landscape.
