z.AI's GLM-5.2 Model Gains US Attention with Low-Cost High Performance

Beijing Zhipu Huazhang Technology (z.AI) released its GLM-5.2 artificial intelligence model on June 16, attracting attention in the US market including from Microsoft as a potential user. The flagship code generation model demonstrates superior performance at approximately one-sixth the API cost of GPT-5.5, with benchmark tests showing 62.1% on SWE-bench Pro compared to GPT-5.5's 58.6%. The model's market entry occurs amid US allegations that Chinese AI companies use 'distillation' techniques to extract proprietary data from American models.

z.AI Releases GLM-5.2 with Superior Benchmark Performance

z.AI released its flagship code generation model GLM-5.2 on June 16, designed for GLM Coding Plan users. According to DataCamp and independent practitioner tests, GLM-5.2 achieved 62.1% on the SWE-bench Pro benchmark test, outperforming GPT-5.5's 58.6%. In the Terminal-Bench 2.1 test, GLM-5.2 reached 81.0% performance. The model features a 1 million token context window option and can output up to 131,072 tokens.

GLM-5.2 Offers One-Sixth API Cost of GPT-5.5

The GLM-5.2 model's application programming interface (API) cost is approximately one-sixth that of GPT-5.5, according to the report by Let's Data Science. Chinese AI models are gaining favor with US customers through their low-cost strategy. The cost advantage comes as Anthropic and OpenAI no longer offer fixed-rate pricing models.

US Companies Accuse Chinese AI Firms of Data Theft in February

Anthropic and OpenAI publicly accused DeepSeek and other Chinese companies in February of using 'distillation' technology to steal proprietary AI data. The US State Department elevated these allegations to the diplomatic level in April, presenting them to governments worldwide. China's embassy responded that the allegations are 'without foundation.'

Microsoft Considers DeepSeek-V4 for Copilot Cowork

Microsoft is reportedly considering adopting a self-hosted modified version of DeepSeek-V4 for its enterprise artificial intelligence tool Copilot Cowork, primarily based on cost considerations. Tesla and SpaceX co-founder Elon Musk predicted that China will achieve frontier-class artificial intelligence capability in the first quarter of 2027.

FAQ

What performance did z.AI's GLM-5.2 achieve on benchmark tests? GLM-5.2 achieved 62.1% on the SWE-bench Pro benchmark test and 81.0% on the Terminal-Bench 2.1 test, according to DataCamp and independent practitioner tests.

When did US AI companies accuse Chinese firms of data theft? Anthropic and OpenAI publicly accused DeepSeek and other Chinese companies in February of using 'distillation' technology to steal proprietary AI data, with the US State Department elevating the allegations in April.