Model positioning
GLM-4-32B-Base-0414 is a base language model with approximately 32 billion parameters in the GLM-4-0414 series, intended for text generation, further training, and local deployment. The official source states that it was pretrained on approximately 15T of high-quality data, including a large amount of reasoning-oriented synthetic data, laying the foundation for subsequent reinforcement learning scaling. (huggingface.co)
Capabilities and uses
- Can serve as a base model for text generation, model research, fine-tuning, and subsequent reinforcement learning.
- The official source lists it separately from GLM-4-32B-0414, the chat model in the same series. The chat model has undergone post-training and has been enhanced for capabilities such as instruction following, coding, and function calling. These post-training capabilities and the chat model's evaluation results should not be treated directly as the performance of this Base version. (github.com)
- The official source provides downloadable weights suitable for self-deployment; the license file in the model repository is the MIT License, whose terms should be followed when using the model. (huggingface.co)
Limitations
- This is a base model, not the chat or deep-reasoning version in the same series; it should be adapted and evaluated for the relevant scenario before direct use in user-facing conversations or complex tasks.
- The official source lists a native context length of 32K, extendable to 128K via YaRN. When the combined input and output may exceed the native length, the official source recommends enabling YaRN; actual results depend on the deployment configuration. (github.com)
- Generated results may be inaccurate or unsuitable for a particular scenario, and important content needs to be verified; this model's separate output limit and knowledge cutoff date have not been verified.
Pricing
As of 2026-11-18, official hosted API input and output prices specifically for GLM-4-32B-Base-0414 had not been verified. The Zhipu domestic pricing page and the Z.ai international pricing page were checked: the domestic page requires JavaScript, so specific prices could not be verified from the content retrieved; GLM-4-32B-0414-128K listed on the international page is a differently named version, and its prices cannot be directly applied to this Base model. Compute, storage, and other costs incurred by self-deployment depend on the resources used. (open.bigmodel.cn)
Release date and official sources
- Release date: 2025-04-14. The official project announcement released the GLM-4-32B-0414 series on that date and listed this Base version in the model list. (github.com)
- Official project announcement and model list
- Official model card and weights
- Official license file