Source URL: https://simonwillison.net/2025/Jul/30/chinese-models/
Source: Simon Willison’s Weblog
Title: The best available open weight LLMs now come from China
Feedly Summary: Something that has become undeniable this month is that the best available open weight models now come from the Chinese AI labs.
I continue to have a lot of love for Mistral, Gemma and Llama but my feeling is that Qwen, Moonshot and Z.ai have positively smoked them over the course of July.
Here’s what came out this month, with links to my notes on each one:
Moonshot Kimi-K2-Instruct – 11th July, 1 trillion parameters
Qwen Qwen3-235B-A22B-Instruct-2507 – 21st July, 235 billion
Qwen Qwen3-Coder-480B-A35B-Instruct – 22nd July, 480 billion
Qwen Qwen3-235B-A22B-Thinking-2507 – 25th July, 235 billion
Z.ai GLM-4.5 and GLM-4.5 Air – 28th July, 355 and 106 billion
Qwen Qwen3-30B-A3B-Instruct-2507 – 29th July, 30 billion
Qwen Qwen3-30B-A3B-Thinking-2507 – 30th July, 30 billion
The only janky license among them is Kimi K2, which uses a non-OSI-compliant modified MIT. Qwen’s models are all Apache 2 and Z.ai’s are MIT.
The larger Chinese models all offer their own APIs and are increasingly available from other providers. I’ve been able to run versions of the Qwen 30B and GLM-4.5 Air 106B models on my own laptop.
I can’t help but wonder if part of the reason for the delay in release of OpenAI’s open weights model comes from a desire to be notably better than this truly impressive lineup of Chinese models.
Tags: open-source, qwen, openai, generative-ai, ai, local-llms, llms
AI Summary and Description: Yes
Summary: The text highlights the recent emergence of powerful open weight models from Chinese AI labs, specifically mentioning several models released in July. It contrasts these with Western offerings, suggesting a growing competitive advantage in model performance and availability. This insight is particularly relevant for professionals in AI security, cloud computing, and generative AI.
Detailed Description: The analysis of the recent developments in AI models reveals a marked shift in the landscape, highlighting key models, their specifications, and licensing types. This information is crucial for industry stakeholders, particularly those invested in AI security and compliance, as it provides insight into the competitive dynamics of model development and the implications for data governance.
– **Key Developments**: The text identifies several significant AI models released in July:
– **Moonshot Kimi-K2-Instruct**: 1 trillion parameters (July 11)
– **Qwen Qwen3 variants**:
– 235 billion parameters (July 21 & 25)
– 480 billion parameters (July 22)
– 30 billion parameters (July 29 & 30)
– **Z.ai GLM-4.5 models**: 355 billion and 106 billion parameters (July 28)
– **Licensing and Compliance**:
– The majority of the listed models have favorable open-source licenses, with Qwen models under Apache 2, while Z.ai models are licensed under MIT.
– Kimi K2’s licensing is noted as “janky,” indicating potential legal or compliance issues, as it uses a non-OSI-compliant modified MIT license. This can be a critical point of concern for organizations relying on these models for secure operations.
– **Availability and Adoption**:
– The larger Chinese models offer APIs and are increasingly accessible through other providers, which raises questions about the competitive landscape and the adoption of foreign technology in enterprise settings.
– The ability to run advanced models like Qwen 30B and GLM-4.5 locally suggests a trend toward democratization of powerful AI tools, allowing for broader experimentation and deployment in various environments.
– **Implications for AI and Cloud Security**:
– The rapid development and release of these models indicate a shift that may force Western AI entities, like OpenAI, to reconsider their strategies regarding model release and capabilities.
– Organizations focusing on AI risk management will need to stay informed about which models to trust, their licensing implications, and the security protocols surrounding their use, including data handling and potential biases in model outputs.
– **Conclusion**: This text is significant for professionals in AI security and cloud computing as it underscores the competitive immediacy of AI development in China and the implications of licensing and adoption trends for compliance and security considerations in the use of generative AI technologies.