Where to download GLM-5.1
Z.ai’s publisher repository lists GLM-5.1 weights on Hugging Face and ModelScope. The Hugging Face pages currently identify these two variants:
- BF16: zai-org/GLM-5.1 on Hugging Face — listed at 1.51 TB.
- FP8: zai-org/GLM-5.1-FP8 on Hugging Face — listed at 756 GB.
Z.ai’s publisher repository labels both variants 744B-A40B: GLM-5.1 is BF16, and GLM-5.1-FP8 is FP8. Hugging Face displays an MIT license on both model pages. Check the license file and the specific repository contents before using the weights.
Download from a terminal
Hugging Face’s huggingface_hub package includes the hf CLI. Install it, then download the repository you want to a chosen local directory:
python -m pip install -U huggingface_hub
hf download zai-org/GLM-5.1 --local-dir ./GLM-5.1
For the FP8 repository, use:
hf download zai-org/GLM-5.1-FP8 --local-dir ./GLM-5.1-FP8
These commands request the repository files; the downloads are very large. Allow at least the listed repository size, plus working space, on the destination drive. The retrieved model listings do not show an access-approval banner, but they do not explicitly guarantee that anonymous downloads will work. If Hugging Face prompts you to authenticate, use hf auth login and follow any access requirements shown on the repository page. See Hugging Face’s CLI guide and download guide for current instructions.
BF16 or FP8: which should you get?
The publisher lists both precision formats, but the supplied documentation does not establish their relative quality, speed, or minimum GPU-memory requirements. Choose based on requirements verified for your hardware and serving framework—not on the precision label alone.
| Repository | Publisher-listed format | Repository size shown on Hugging Face |
|---|---|---|
zai-org/GLM-5.1 |
744B-A40B, BF16 | 1.51 TB |
zai-org/GLM-5.1-FP8 |
744B-A40B, FP8 | 756 GB |
The FP8 repository is smaller in the listing, but that alone does not establish that it will fit a particular machine or prove a quality or performance trade-off. Z.ai’s repository names SGLang, vLLM, Transformers, KTransformers, and Unsloth for GLM-5.3 and earlier GLM-5 models; it does not provide a GLM-5.1-specific hardware matrix in the referenced text.
Prefer a production API?
Z.ai’s model materials point to API services for GLM-5.1, and its standard API pricing page lists the model. The general chat-completions documentation gives the endpoint format and says to use a Standard API Key with bearer authentication. Do not substitute a GLM Coding Plan key or endpoint: Z.ai says those credentials are managed separately, and Coding Plan requests for GLM-5.1 are automatically routed to GLM-5.3.
The following is a schema-based example, not a model-specific request confirmed against your account. The general Z.ai chat-completions example uses this endpoint and request structure, but its sample model is GLM-5.3. Before using this for production, confirm that glm-5.1 is enabled and accepted by your standard API account.
curl --request POST \
--url https://api.z.ai/api/paas/v4/chat/completions \
--header "Authorization: Bearer $ZAI_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"model": "glm-5.1",
"messages": [
{"role": "user", "content": "Summarize the purpose of this repository."}
],
"stream": false
}'
Z.ai’s pricing table lists GLM-5.1 prices in USD per 1 million tokens: $1.40 input, $0.26 cached input, limited-time free cached-input storage, and $4.40 output. The free promotional entry is for cached-input storage—not a free-input rate. Confirm current prices and account terms before budgeting.
Quick choice: use the hosted API if you want to avoid downloading terabytes of weights and managing local inference. Choose a weight repository if you need the files for your own deployment and have verified your hardware and serving framework can handle the selected variant.
Avoid unverified mirror listings
The captured SourceForge listing whose URL identifies it as a mirror does not show downloadable files or a working download path. Use the publisher-linked repositories above rather than treating that listing as proof of an official download.
Sources
Z.ai GLM-5 publisher repository
Z.ai GLM-5.1-FP8 on Hugging Face
Hugging Face Hub download guide
Z.ai Chat Completion API reference