High CPU usage on macOS after updating Codex VS Code extension to 0.50.x+ (0.4.49 works fine)

Open 💬 18 comments Opened Dec 13, 2025 by voltCore
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

Codex 0.4.50+

What subscription do you have?

ChatGPT Plus

Which model were you using?

GPT-5.2 (also reproduced with GPT-5.1)

What platform is your computer?

MacBook with Apple Silicon (M2) MacOS Tahoe 26.2

What issue are you seeing?

After updating the Codex VS Code extension to version 0.50.x +, CPU usage increases significantly on macOS only when a prompt is sent and Codex starts processing it.

While Codex is actively thinking or generating a response, a single process — "Code Helper (Plagin)" — consumes excessive CPU (up to 300–350%) on Apple Silicon.

<img width="932" height="70" alt="Image" src="https://github.com/user-attachments/assets/9272481e-f484-42f5-ace6-4e6454b88cd4" />

The editor remains responsive, but system performance degrades noticeably during prompt processing.

Downgrading the extension to version 0.4.49 resolves the issue completely, with CPU usage remaining at normal levels during prompt execution.

What steps can reproduce the bug?

  1. Install Codex VS Code extension version 0.50.x +
  2. Launch VS Code on macOS (Apple Silicon)
  3. Open any project (JavaScript, Python, or mixed)
  4. Open the Codex chat panel and switch agent (Full access) GPT-5.1-Codex-Max High
  5. Enter any prompt and send it
  6. While Codex is processing the prompt, observe system CPU usage in Activity Monitor

Result:
During prompt processing, the "Code Helper (Plugin)" process spikes to 300–350% CPU and my laptop freezing

Workaround:
Downgrading the extension to version 0.4.49 eliminates the high CPU usage during prompt processing.

What is the expected behavior?

During prompt processing, Codex should utilize system resources efficiently and keep CPU usage within reasonable limits.

CPU usage should scale proportionally to the complexity of the prompt and should not saturate multiple CPU cores on Apple Silicon for simple or medium-sized requests.

Additional information

_No response_

View original on GitHub ↗

18 Comments

github-actions[bot] contributor · 7 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #7456
  • #7822

Powered by Codex Action

etraut-openai contributor · 7 months ago

Does this happen for all sessions, or is it just in specific sessions? In your repro steps, you said "enter any prompt". Does it occur only for prompts that result in tool execution? For example, if your prompt is simply "Hi!", does the problem occur?

Please use the /feedback command to upload logs for a session where this occurs and post the thread ID here.

michaelsthr · 7 months ago

I have the same issue since the update. CPU usage spikes extremly high after prompting anything

etraut-openai contributor · 7 months ago

@michaelsthr, please use the /feedback command to upload logs for a session where this occurs and post the thread ID here.

andrewlm6 · 6 months ago

Also seeing this issue, thread id below. Seems to be worse on long threads, backing out of it fixes it, even though Codex continues running. M5 Macbook

019bdec8-8c64-7b80-aca9-6123a7bb2268

Update: Latest version fixed this issue.

doggie1989 · 4 months ago

50% cpu ... m4max with 128g ram

ManuCatZe · 4 months ago

Same Problem here

tom10271 · 3 months ago

Same here, I have to shut down VS Code from time to time to save battery power after getting Codex response. CPU usage stays low before I ask anything for Codex

ljantos · 3 months ago

Same here, with the same strategy of @tom10271

vidstige · 3 months ago

VS Code burning through massive amount of CPU. Almost unusable as the battery drains so fast. In the Process Explorer the "window" process name that using the CPU. If I kill that process the Codex panel goes completely gray.

jinranh · 3 months ago

Same here

ssinha2103 · 3 months ago

Is there any suggestion to fix it urgently ?

okanakdogan26 · 3 months ago

experiencing the same

Hose9817 · 3 months ago

I have the same problem

isokasakissoja · 3 months ago

Same problem. A prompt doesn't have to even be processing, you just need to have Codex tab opened (switching to a different chat tab reduces the power use instantly).

ilya-evseev · 3 months ago

Same issue. Codex causes my MacBook Pro M4 overheating.
High CPU load on window process starts right after Codex finishes processing the request and makes changes to the project files. See the screenshot. If I kill that overloading CPU window process, Codex extension goes gray.
Makes it impossible to work with Codex extension in vscode.

<img width="666" height="499" alt="Image" src="https://github.com/user-attachments/assets/9c1a4e8b-2984-4ac6-8b3b-861ed94d92d7" />

shadow-identity · 1 month ago

Still reproducible here on a current VS Code extension build.

Environment:

  • macOS 26.5 (25F71), arm64
  • VS Code 1.117.0 arm64
  • OpenAI ChatGPT / Codex extension: openai.chatgpt@26.519.32039
  • Two VS Code windows open, each with its own Codex extension host

Observed:

  • Two Code Helper (Plugin) processes reached roughly 333-343% CPU after the extension hosts were respawned.
  • The child codex app-server processes stayed at 0.0% CPU during the spike, so this seems to be in the VS Code extension host, not in the app-server child.
  • Both window logs were growing and repeatedly included:

[IpcClient] Received broadcast but no handler is configured method=thread-stream-state-changed

  • Killing the hot helper processes is only temporary because VS Code respawns the extension hosts.

This appears to be the same family of issue as #23266.

tom10271 · 1 month ago

I personally do not use VS Code to code and I switched to use Codex's desktop app. So far I do not have the high CPU problem there. For VS Code users you can just use the Codex app but not the VS Code plugin. This issue has been persisted so quite a while, frankly speaking it would be probably fixed in 2027. Get a life