Zhipu, a Beijing-based AI startup spun out of Tsinghua University, said developers sharply increased use of its open-weight GLM 5.2 model after its June release, as companies sought cheaper AI tools for coding, planning, and testing.

The model’s weights are MIT-licensed, allowing companies to download, self-host, fine-tune, and deploy it in air-gapped environments.

Running the roughly 744 billion-parameter mixture-of-experts system still requires substantial graphics processing unit memory.

Zhipu said GLM 5.2 was competitive with Anthropic’s Claude Opus on agentic and coding tasks at a lower cost.

The comparison appeared to rely on company claims and early community feedback rather than an independent benchmark.

OpenRouter said token traffic for GLM 5.2 rose faster than after DeepSeek’s V4 launch in April, while Harvey co-founder Gabe Pereyra said the model was competitive with some closed-source systems.

The launch is notable because Zhipu was China’s third-largest large language model player in 2024 and raised 2.5 billion yuan (US$368 million) from backers including Alibaba and Tencent in 2023.

Chinese open models are also seeing demand from cost-sensitive enterprise users, while US restrictions and company geoblocking limit access to some proprietary systems in China.

🔗 Source: CNBC