Describe the Bug
Bug: GLM 5.2 (high / max) does not invoke Cursor IDE tools — only replies in chat
Summary
When using the GLM 5.2 model in high and max (reasoning) modes, the model does not initiate tool calls within the Cursor IDE. Instead of executing commands via shell, reading files, searching, etc., the model only generates a text response in chat, describing what it intends to do without actually invoking the tools.
Environment
OS: macOS (darwin 25.5.0)
IDE: Cursor
Model: GLM 5.2 (high / max reasoning)
Comparison: other models (e.g., Claude Sonnet, GPT-5.x) correctly invoke tools under the same conditions.
Reproduction steps
Open Cursor IDE, start a new chat with the GLM 5.2 model (high or max).
Send a request that requires a tool call, for example:
“Use the Cursor CLI and ping any model from there”
“Read the package.json file and tell me what dependencies are installed”
“Run git status in the terminal”
Observe the model’s response.
Expected behavior
The model should invoke the corresponding IDE tools (Shell, Read, Grep, etc.) and return the result of their execution, as other models do under the same conditions.
Actual behavior
The model only replies with text in chat. In the first response it describes the intent (“let me check…”) but does not invoke the tool. Only after the user explicitly points it out again (“what’s the problem, why did you reply in chat instead of calling the tool”) does the model start invoking tools.
Impact
GLM 5.2 (high/max) cannot be used for agentic tasks in Cursor: code edits, command execution, repository search, file reading — none of these run until a repeated “nudge”.
Increased latency and token consumption: the user is forced to repeat the request to make the model invoke a tool.
Behavior is unpredictable: after one “nudge” the model starts working correctly, but in a new session the bug reproduces again.
Hypothesis (for investigation only, not for fixing)
The issue may be related to how the reasoning variants of GLM 5.2 (high/max) process the system prompt describing available tools: the model “thinks” through the response but never reaches the step of emitting tool-call tokens, or the emission happens but is not parsed correctly by the Cursor frontend. In non-reasoning modes (low/medium) the bug may not reproduce — worth checking.
Reproducibility
Reproduces consistently in a new session with GLM 5.2 high/max.
Does not reproduce with other models in the same session/configuration.
Steps to Reproduce
Open Cursor IDE, start a new chat with the GLM 5.2 model (high or max).
Send a request that requires a tool call, for example:
“Use the Cursor CLI and ping any model from there”
“Read the package.json file and tell me what dependencies are installed”
“Run git status in the terminal”
Observe the model’s response.
Expected Behavior
The model should invoke the corresponding IDE tools (Shell, Read, Grep, etc.) and return the result of their execution, as other models do under the same conditions.
Operating System
MacOS
Version Information
Version: 3.10.20 (Universal)
VS Code Extension API: 1.125.0
Commit: 23b9fb205fe595ea2be29da7214e19762d037fc0
Date: 2026-07-07T07:03:33.071Z
Layout: IDE
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Darwin arm64 25.5.0
For AI issues: which model did you use?
z.ai glm 5.2 HIGH / MAX
Does this stop you from using Cursor
No - Cursor works, but with this issue