Chinese AI startup Z.ai, formerly known as Zhipu AI, released the GLM-5.2 open-weights model on June 16, featuring a 753-billion-parameter system. The model outperforms OpenAI’s GPT-5.5 on several long-horizon coding benchmarks and costs roughly one-sixth as much to operate via API.
GLM-5.2, which is available under an MIT open-source license on Hugging Face, scored 62.1 on SWE-bench Pro, compared to GPT-5.5’s score of 58.6. Additionally, it achieved a score of 81.0 on Terminal-Bench 2.1, becoming the first open-weights model to surpass 80% on that benchmark. However, on Terminal-Bench, it trails behind Claude Opus 4.8 and GPT-5.5, which scored 85.0 and 84.0, respectively.
The API pricing for GLM-5.2 is set at $1.40 per million input tokens and $4.40 per million output tokens, with cached input costing around $0.26 per million. This pricing structure allows GLM-5.2 to be approximately one-sixth the cost of GPT-5.5 for similar coding tasks, according to VentureBeat.
GLM-5.2 supports a one-million-token context window and may generate up to 131,072 output tokens per response. The model is accessible through Z.ai’s API, Cloudflare Workers AI, and over 20 third-party coding environments.
Z.ai initially provided GLM-5.2 to paid coding plan subscribers on June 13 before the full open-weights release. The model scored 99.2 on the AIME 2026 benchmark, outperforming both GPT-5.5 and Claude Opus 4.8.
The introduction of GLM-5.2 aligns with a trend of competitive open-source models emerging from Chinese AI labs. It is the fourth model in the GLM-5 series, following GLM-5, GLM-5-Turbo, and GLM-5.1. Z.ai founder Jie Tang stated that the release is partly a response to recent restrictions on U.S.-based AI services.
