Zhipu GLM-5.2 Review

Zhipu, now trading as Z.ai, put GLM-5.2 out on 13 June 2026, and it is the clearest sign yet that the open weight gap has all but closed. The architecture is a 744 billion parameter mixture of experts with roughly 40 billion active per token, a full one million token context and selectable reasoning modes. On independent long horizon coding tests it matches or beats OpenAI's GPT-5.5, and it does so at about one sixth of the running cost.

Pricing is aggressive from either direction. Through the API it is around 1.40 dollars per million input tokens and 4.40 dollars per million output, while the GLM Coding Plan starts at 18 dollars a month. If you would rather host it yourself, the weights sit on Hugging Face and ModelScope under a permissive MIT licence, so there is no gate and no per seat tax.

The reasons to pause are the familiar ones for a frontier Chinese model. Data residency, compliance and the politics of where your prompts are processed all matter for some buyers, and the tooling ecosystem around it, while growing fast, is still thinner than the Western incumbents. For coders and builders who care about capability per dollar and want the option to run everything in house, GLM-5.2 is one of the strongest picks of the year.