MiniMax M2.7 Release Intensifies China AI Model Race
Chinese AI startup MiniMax released the weights of its M2.7 model on April 12, adding momentum to a growing wave of open-weight models from Chinese labs. As a result, the competitive landscape continues to shift ahead of DeepSeek’s expected V4 launch. The model, first introduced in March, delivers strong benchmark performance across coding and project execution tasks. Consequently, it positions itself among the leading open-weight models currently available.
Performance Gains and Licensing Debate
MiniMax M2.7 achieved 56.22 percent on SWE-Pro and 55.6 percent on VIBE-Pro, which places it near top-tier coding systems. In addition, it reached an ELO score of 1,495 on GDPval-AA, ranking highest among open-weight models. The weights are now accessible on Hugging Face and supported on NVIDIA platforms, which broadens developer access.
However, the release also triggered debate within the developer community. The license restricts commercial use without prior permission, which raises questions about its classification as open source. At the same time, the model introduces a “self-evolution” development method, where it contributes to its own training cycle. Therefore, it reflects both technical progress and ongoing licensing tensions.
China’s Expanding AI Ecosystem and DeepSeek Focus
Meanwhile, other Chinese labs continue to accelerate development. Zhipu AI released GLM-5.1 under an MIT license, which supports extended autonomous task execution. In contrast, Alibaba introduced Qwen 3.6 Plus as a proprietary API model, signaling a shift in strategy.
At the same time, attention is turning toward DeepSeek’s upcoming V4 model. Reports indicate the system will run on Huawei Ascend chips, which support China’s push for domestic hardware. Additionally, the model is expected to feature a Mixture-of-Experts architecture with a large context window and multimodal capabilities. Although delays have occurred, recent testing activity suggests a launch may be imminent.
