Run GLM 5.2 Locally
Use zai-org/GLM-5.2 with vLLM, SGLang, Transformers, KTransformers, or Unsloth.
Supported Local Paths
vLLM
OpenAI-compatible API serving
Production and private API deployments
SGLang
High-throughput inference server
Agentic coding and batch workloads
Transformers
Python research workflow
Scripts, notebooks, and experimentation
KTransformers
Optimized local inference
Workstation experiments
Unsloth
Fine-tuning workflow
Adapter training and evaluation
Step-by-Step Setup
1
Get the model
Use the official HuggingFace repository for the current open weights.
huggingface-cli download zai-org/GLM-5.2 --local-dir ./GLM-5.2
2
Serve with vLLM
vLLM is one of the supported local serving frameworks for GLM 5.2.
pip install vllm vllm serve zai-org/GLM-5.2 --served-model-name glm-5.2
3
Serve with SGLang
SGLang is also listed as a supported GLM 5.2 deployment path.
pip install sglang python -m sglang.launch_server --model-path zai-org/GLM-5.2 --served-model-name glm-5.2
4
Call the local server
Most serving frameworks expose an OpenAI-compatible chat completions API.
curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
"model": "glm-5.2",
"messages": [{"role": "user", "content": "Write a Python hello world"}]
}'Use a Local Server with OpenCode
Once your local OpenAI-compatible server is running, point OpenCode at it:
# In your OpenCode config (opencode.json)
{
"model": {
"provider": "openai",
"name": "glm-5.2",
"baseURL": "http://localhost:8000/v1",
"apiKey": "local"
}
}Tips for Best Performance
- Use vLLM or SGLang for API serving rather than old GLM4 Ollama examples.
- Plan capacity around the 1M-token context. Full-context runs need much more memory.
- Use the official HuggingFace repository: zai-org/GLM-5.2.
- For production, compare local serving cost against Z.ai or OpenRouter before committing.