Z.ai released GLM-5, which is positioned as a large model for complex system engineering and long-term "Agentic" tasks, and simultaneously provides online experience and third-party call portals. According to the official disclosure, compared with GLM-4.5, GLM-5 has improved in scale and training data, with parameters expanded from 355B to 744B, activation parameters increased from 32B to 40B, and pre-training data increased from 23T to 28.5T tokens.
GLM-5 weights have been published on Hugging Face and marked with MIT license, and call pages are available on platforms such as OpenRouter; Z.ai subscription page shows that GLM-5 can be used in a variety of AI programming toolchains. Subscription-related restrictions include that different tiers may be opened in batches, and that new models may consume more quota; At the same time, external reports show that Zhipu-related subscription products have price adjustment information due to changes in demand, and users need to pay attention to the risk of fluctuations in actual billing and usage costs.
FAQs
Q: What type of model is GLM-5?
A: GLM-5 is the flagship open-source large model released by Z.ai, focusing on complex system engineering and long-range agent workflows.
Q: What are the changes in the parameter scale and training data of GLM-5?
A: Public information says that its total parameters are 744B, activation is 40B, and the pre-training data is 28.5T tokens, and it compares the previous generation of 355B/32B and 23T.
Q: How do ordinary users experience GLM-5?
A: It can be used in the chat portal of the Z.ai, or it can be called through API on aggregation platforms such as OpenRouter.
Q: What is the opening rhythm and considerations of the GLM Coding Plan?
A: The public prompt is available for Max users first, and other gears may be opened on a rolling basis; You should also pay attention to the quota consumption and changes in subscription fees.
