Codex app-server stuck at 100% CPU on macOS Apple Silicon when opening VS Code ChatGPT extension
What version of Codex is running?
Tested both: openai.chatgpt 0.4.46 (stable release) • openai.chatgpt 0.5.47 (pre-release, Mac Silicon build) Both versions reproduce the same issue
What subscription do you have?
ChatGPT Plus
Which model were you using?
GPT-5.1-Codex_Max, GPT 5.1
What platform is your computer?
Darwin 23.6.0 arm64 arm
What issue are you seeing?
When I open the ChatGPT / Codex panel in VS Code on macOS Apple Silicon, the codex app-server process immediately rises to ~100% CPU and stays there for several minutes. During this time:
- The ChatGPT panel takes a very long time to initialize (often tens of seconds or more).
- About one out of every three prompt submissions gets stuck with a spinner for several minutes before finally responding.
- CPU usage remains very high throughout these delays.
After some minutes, the panel eventually loads and prompt submissions eventually complete, but the codex app-server process never stabilizes to low CPU usage — it continues running near 100% until VS Code is closed or the process is manually killed.
I tested both:
- Stable version:
openai.chatgpt0.4.46 - Pre-release version:
openai.chatgpt0.5.47 (Mac Silicon VSIX)
Both show exactly the same behavior.
What steps can reproduce the bug?
- Close all VS Code windows.
- Clean previous processes:
``bash``
pkill -f "codex app-server"
ps aux | grep codex
(Only the grep line should remain.)
- Open VS Code with a single window (empty workspace or very small project).
- Open the ChatGPT / Codex sidebar.
- Observe that:
• a single codex app-server process appears,
• CPU instantly jumps to ~100%,
• the panel takes a long time to finish initializing,
- Submit prompts:
• some are processed normally after a delay,
• roughly 1 out of every 3 prompts gets stuck with the spinner for several minutes before eventually completing.
What is the expected behavior?
The codex app-server process should briefly spike during startup and then settle to low CPU usage. The ChatGPT / Codex panel should load quickly, and prompt submissions should respond promptly and consistently.
Instead:
- panel initialization takes an unusually long time,
- prompt submissions often get stuck for minutes,
- and CPU use stays near 100% for the entire time the extension is active.
Additional information
System details:
- Mac Studio with Apple M2, 32 GB RAM
- Apple Silicon (ARM64)
- macOS 14.8 Sonoma
- VS Code 1.106.3
- ChatGPT / Codex extension:
- Stable: 0.4.46
- Pre-release: 0.5.47 (Mac Silicon VSIX)
I am not using Codex CLI outside the extension. No MCP servers configured. The issue happens even with a single VS Code window and an empty workspace.
If needed, I can provide the “Output → Codex” logs or Activity Monitor traces.
This issue has 6 comments on GitHub. Read the full discussion on GitHub ↗