Zhipu AI
Core Facts
The GLM series models from Zhipu (Z.ai) were jointly developed by Tsinghua University and Zhipu AI
@ai-sdk/zai is built on the @ai-sdk/openai-compatible compatibility layer and only requires configuring Zhipu-specific endpoint addresses and parameters.
Zhipu released GLM-5.3, claiming its coding capability improved by 50% compared to GLM-5.2.
Timeline (last 90 days)
Zhipu's GLM-5.3 model is now available on Cursor.
Zhipu makes heavy use of interns, with more interns than full-time employees. A team of 10 full-time employees paired with 20 interns is considered normal at model vendors.
In the VALS AI legal evaluation, GLM-Legal took first place among open-weight models on both the Harvey Lab and Legal Bench leaderboards, with a 19.2% all-tasks-passed rate and 86.23% accuracy.
Zhipu launched GLM-Legal, an open-source large model dedicated to law, for practical work such as contract review, evidence linking and cross-border compliance research.
In July of this year, the Ministry of Commerce consulted companies including Alibaba, ByteDance and Zhipu about possible control measures, with discussions including tightening restrictions on overseas downloads of model weights while allowing overseas customers to continue using hosted Chinese models.
White-hat security researcher Gu He (Yuki Chen) has formally joined Zhipu's GLM foundation model team, and the 23 Windows vulnerabilities his team mined using GLM models have all been confirmed and acknowledged by Microsoft MSRC.
Many multimodal large models on the market (such as the Zhipu series, the Tongyi Qianwen series, and the GPT series) support multimodal input, but their output does not support multimodality.
Recently, the Agent and Coding scores of the official DeepSeek version, Kimi K3, and Zhipu's latest model on well-known foreign testing platforms have come very close to the level of models such as the OpenAI GPT-5.5/5.6 series and Anthropic Opus 4.8.
Large model vendors such as Zhipu and MiniMax have opened stores on Tmall, selling API call quotas (tokens) as goods during Double 11.
Zhipu is giving some subscription users gray-scale access to GLM-5.3 High Speed, and users can randomly obtain trial access in Zcode.
40 more timeline events
All Facts (20)
The GLM series models from Zhipu (Z.ai) were jointly developed by Tsinghua University and Zhipu AI
75%Verified@ai-sdk/zai is built on the @ai-sdk/openai-compatible compatibility layer and only requires configuring Zhipu-specific endpoint addresses and parameters.
75%VerifiedZhipu released GLM-5.3, claiming its coding capability improved by 50% compared to GLM-5.2.
65%Unverified国内合规的大模型替代方案包括百度文心一言、阿里通义千问、智谱ChatGLM等
90%Unverified阿里的Qwen-VL和智谱的CogVLM/GLM-4V是中文领域代表性的多模态模型
90%UnverifiedZhipu's GLM-5.3 model is now available on Cursor.
50%UnverifiedZhipu makes heavy use of interns, with more interns than full-time employees. A team of 10 full-time employees paired with 20 interns is considered normal at model vendors.
50%UnverifiedIn the VALS AI legal evaluation, GLM-Legal took first place among open-weight models on both the Harvey Lab and Legal Bench leaderboards, with a 19.2% all-tasks-passed rate and 86.23% accuracy.
50%UnverifiedMany multimodal large models on the market (such as the Zhipu series, the Tongyi Qianwen series, and the GPT series) support multimodal input, but their output does not support multimodality.
50%UnverifiedRecently, the Agent and Coding scores of the official DeepSeek version, Kimi K3, and Zhipu's latest model on well-known foreign testing platforms have come very close to the level of models such as the OpenAI GPT-5.5/5.6 series and Anthropic Opus 4.8.
50%UnverifiedZhipu's GLM-5 identified itself as Gemini made by Google in a Spanish-language scenario, and after being misled by a suggestive question, identified itself as Claude made by Anthropic.
50%UnverifiedZhipu's API gross margin in the first half of the year was only 25%, while Moonshot AI's API gross margin last year was approximately 45%.
50%UnverifiedDeepSeek's overall gross margin in July was approximately 45%, below Anthropic's 60%; OpenAI's gross margin is approximately 39%, Zhipu AI's is 26%, and MiniMax's is 18%.
50%UnverifiedThe open-source model GLM 5.3 Flash launched by Zhipu topped the Cyber Index leaderboard on Artificial Analysis.
50%UnverifiedCAICT and Lüméng Technology evaluated and confirmed the rectification measures taken for Zhipu's ZCode.
50%UnverifiedCCSwitch has built-in support for mainstream model providers such as DeepSeek and Zhipu, and also supports custom configurations and third-party relay stations.
50%UnverifiedAn Anthropic red-team evaluation called out Zhipu GLM-5.3, stating it can autonomously write complete exploit code, with a success rate of up to 100% in bypassing its own defenses, and that its safety is lower than Claude models at the same tier.
50%UnverifiedCC Switch has most mainstream model providers on the market built in, such as DeepSeek and Zhipu, and providers not on the list can be configured using custom settings.
50%UnverifiedIn the open-weight large model track, Chinese companies are leading, with the main players including DeepSeek, Zhipu AI, and Alibaba Qwen, while France's Mistral and NVIDIA are also competing in the same track.
50%UnverifiedA large number of players including Mistral, Cohere, Anthropic, Meta (LLaMA series), DeepSeek, and Zhipu have entered the frontier model market.
50%