If DeepSeek is the cheapest way to give OpenClaw a brain, GLM is the most predictable. Z.AI’s GLM models come with a flat monthly coding plan, which means an always-on assistant costs the same whether it has a quiet week or a busy one. For anyone who’d rather not watch a token meter, that fixed cost is the whole appeal.
This guide sets up OpenClaw with GLM: the config, the plan-versus-pay-per-token decision, key handling, and the usage caps worth knowing about. It assumes OpenClaw is installed — if not, see the Windows and WSL install guide first.
Why GLM for OpenClaw
The reason to pick GLM over a pure pay-per-use model is billing. A flat coding plan turns a variable, slightly nerve-wracking cost into a line item you forget about. That matters more with OpenClaw than with a chat tool, because an assistant on a schedule works in the background — and on a flat plan, background work doesn’t move the bill.
GLM vs DeepSeek for OpenClaw
| GLM (coding plan) | |
| Billing | Flat monthly subscription |
| Best for | Heavy, daily, always-on use |
| Cost certainty | Fixed |
| Caps | Rolling-window usage limits |
If your use is light or unpredictable, DeepSeek is probably cheaper. If you lean on OpenClaw all day, GLM’s plan usually wins.
Step 1: get a GLM key and pick a billing model
Sign up with Z.AI, decide between the coding plan and pay-per-token, and generate an API key. Note the base URL and current model names from the Z.AI developer docs. The plan-versus-token choice is worth a moment — estimate how much you’ll actually run OpenClaw before committing, since the plan only pays off above a certain volume.
Step 2: store the key in an environment variable
export GLM_API_KEY="your-z-ai-key"Add it to your shell profile (~/.bashrc or ~/.zshrc) so it survives new sessions. As always, keep it out of any config file you might share.
Step 3: configure OpenClaw to use GLM
Set GLM as the model provider in OpenClaw’s config, using Z.AI’s base URL and a GLM model name, with the key pulled from the environment. Config shapes change between versions, so confirm against the OpenClaw repo; the values you’re providing stay the same:
{
"model": {
"provider": "openai-compatible",
"base_url": "https://api.z.ai/api/paas/v4",
"api_key_env": "GLM_API_KEY",
"model": "glm-5"
}
}Confirm the exact base URL and model name in Z.AI’s docs — those are the two values most likely to differ from this example. Everything else is OpenClaw’s standard provider setup.
Step 4: test it
Start OpenClaw and hand it a task that uses a capability, not just chat:
Read the files in this folder and tell me which ones changed most recently.A sensible answer means GLM is connected and the assistant can act. If the model talks but can’t reach files, the model is fine and you’re looking at a skill or permission, not GLM.
OpenClaw + GLM checklist
- Z.AI account created; coding plan vs pay-per-token decided
- API key generated and set as an environment variable
- OpenClaw provider set to Z.AI's base URL with a GLM model
- Per-window usage cap checked if you're on the plan
- Tested with a task that touches files
When GLM is the right call
Reach for GLM when predictability matters more than squeezing the last cent. A flat plan suits someone who runs OpenClaw as a daily driver — scheduled jobs, frequent tasks, an assistant they actually lean on. If instead you fire it up occasionally, the plan’s monthly cost is dead weight and DeepSeek’s pay-per-use is the better fit.
Wrapping up
Setting OpenClaw up on GLM is the same short config as any provider — base URL, key from an environment variable, model name — with one extra decision up front: the coding plan versus pay-per-token. Choose the plan if you run OpenClaw hard every day and want a fixed cost, and mind the rolling-window cap so a busy day doesn’t surprise you.
For the alternatives, compare against DeepSeek, or skip API costs entirely with a local Ollama setup.
