MiniMax released the weights of its MiniMax M2.7 model on April 12, 2023, amid a wave of open-weight releases from Chinese labs, significantly impacting the competitive AI landscape.
This release aligns with a broader trend in China, as various companies rush to unveil their advancements. MiniMax M2.7 was first announced in March and scored 56.22% on the SWE-Pro benchmark, matching the results of GPT-5.3-Codex. Additionally, it achieved 55.6% on the VIBE-Pro benchmark, which is nearly on par with Opus 4.6 for project delivery in web, mobile, and simulation tasks. The model also obtained the highest ELO rating of 1,495 on the GDPval-AA assessment among open-source models.
The weights of MiniMax M2.7 are now available on Hugging Face and supported on NVIDIA platforms. However, the release has faced criticism due to licensing restrictions that prohibit commercial use without prior written permission, which some community members argue undermines its open-source classification. MiniMax characterized M2.7 as the first model to participate in its own development cycle through a process they call “self-evolution.”
MiniMax’s release follows Zhipu AI’s open-sourcing of its GLM-5.1 model on April 7, which operates under the MIT license. This model, containing 754 billion parameters, can execute engineering tasks autonomously for up to eight hours. Conversely, Alibaba launched its proprietary API model, Qwen 3.6 Plus, on April 2, sparking speculation on its commitment to open-source practices. Qwen 3.6 Plus was still free on OpenRouter as of April 11, but its long-term availability remains uncertain.
Anticipation is building for DeepSeek’s V4 model, reported to launch in late April. DeepSeek founder Liang Wenfeng indicated that the model will run on Huawei’s Ascend chips, underscoring China’s semiconductor self-sufficiency efforts. Expected to feature roughly one trillion parameters and utilize Mixture-of-Experts architecture with a one-million-token context window, V4 is also anticipated to include multiple variants, such as a Vision mode for multimodal support. Despite experiencing two delays, early stress testing suggests a public launch may be imminent.
