GLM Models Provider

Z.AI GLM models as a VS Code Language Model Chat Provider for the Coding Plan.

Maintained fork — the original extension by DenizhanDaklr is no longer maintained. This fork continues development with support for the latest GLM-5.3 models.

Why only GLM-5.3 and GLM-5.3-Flash?

Z.AI has moved the GLM Coding Plan to GLM-5.3 and GLM-5.3-Flash. Per the official plan overview, requests for GLM-5.2/GLM-5.1 are automatically routed to GLM-5.3, and requests for GLM-4.7 are routed to GLM-5.3-Flash -- the legacy models no longer exist as distinct options. Keeping them in the picker would only add clutter:

• GLM-5.3 is the current flagship -- same price as GLM-5.2, with significantly stronger coding and long-horizon performance, plus a full 1M-token context.
• GLM-5.3-Flash outperforms GLM-5.2 while costing roughly 20x less, and it is natively multimodal (image input).

Text Models

Vision Models

Commands

• GLM: Set API Key -- Store your Z.AI API key in VS Code secrets
• GLM: Clear API Key -- Remove the stored API key
• GLM: Manage Provider -- Open provider management options
• GLM: Set Thinking Effort -- Choose reasoning effort (Auto, Low, High, Max)
• GLM: Set Temperature -- Choose a temperature preset or enter a custom value (0.0 - 1.0)

Thinking Mode

GLM-5.3 models always use reasoning/thinking -- it cannot be disabled. You can control the reasoning depth with reasoning_effort.

Run GLM: Set Thinking Effort from the Command Palette to choose between:

• Auto -- Use the API default effort level (max)
• Low -- Lightweight reasoning, faster responses
• High -- Enhanced reasoning
• Max -- Deep reasoning, best for complex tasks (API default)

The selected value is persisted in your VS Code settings under glm-models-provider.defaultThinkingMode.

Sampling Defaults

Per the official GLM-5.3 docs, Z.AI recommends temperature: 1.0 and top_p: 0.95 for GLM-5.3 models, so the extension applies them by default:

• The model picker's temperature setting defaults to Max (1.0). You can still pick a preset or custom value per model, or run GLM: Set Temperature.
• Every request sends top_p: 0.95 (configurable via glm-models-provider.topP).
• When a request streams with tools, tool_stream: true is sent alongside stream: true.

How to Use

1. Open the model picker in VS Code's Chat view and choose Manage Models…
2. Select Z.AI GLM and enter your API key when prompted
3. Pick GLM-5.3 or GLM-5.3 Flash from the model picker

Step 1 is what makes the models appear. VS Code passes the key it collects there to the provider, and the provider lists no models until it arrives.

GLM: Set API Key stores a key in the extension's own secret storage instead. That copy backs Test Connection and plan-quota polling when no key has been entered through the picker — it does not, on its own, put models in the picker.

License

MIT (c) Denizhan Dakilir