High GPU usage (70–90%) on macOS with Codex app

[tavaresgmg]

What version of Codex is running?

latest (codex desktop app launch 2 Feb 2026)

What subscription do you have?

ChatGPT Pro

Which model were you using?

gpt-5.2 codex high

What platform is your computer?

Darwin 25.2.0 arm64 arm

What terminal emulator and version are you using (if applicable)?

Codex desktop app (no terminal)

What issue are you seeing?

When running the Codex desktop app with the gpt-5.2 codex high model on macOS (Darwin 25.2.0 arm64), the application consumes between 70% and 90% of my MacBook Pro M1 Pro's GPU during normal operation. This results in heavy energy use and high temperatures. For a language-based coding assistant, such intense use of the local GPU is unexpected compared to other applications.

What steps can reproduce the bug?

1. On a MacBook Pro M1 Pro (32 GB RAM, 1 TB) running macOS 14.x (Darwin 25.2.0), launch the Codex desktop application using the gpt-5.2 codex high model with a ChatGPT Pro subscription.
2. Use the app normally to generate or edit code; it does not need to be idle.
3. Open Activity Monitor and observe the GPU usage of the Codex/Electron process while the app is performing operations.
4. GPU usage stays between 70% and 90% during active use.

What is the expected behavior?

The Codex application should use minimal GPU and CPU resources because most processing occurs on OpenAI's servers. During regular coding assistance, the client should use less than 10% of the GPU. Consuming 70‑90% of the GPU is not expected for a text-based code assistant.

Additional information

Machine: 2021 14-inch MacBook Pro with M1 Pro (10-core CPU, 16-core GPU), 32 GB unified memory, 1 TB SSD. App: latest Codex desktop release (launched Feb 2 2026), ChatGPT Pro subscription, gpt-5.2 codex high model. The high GPU utilization persists after reboots and does not occur with other apps. Please let me know if additional logs or flags would help.

[etraut-openai]

Thanks for the bug report. I'm not aware of any other reports of this behavior, so we'll need to figure out what's different about your computer, configuration, or usage.

Is there anything special or noteworthy about the coding tasks or tools that the agent is running on your system? If you can repro this on demand, could you use the /feedback slash command to upload your logs and session details and then post your thread ID here?

[etraut-openai]

Are you seeing especially large diffs during your session? We've had some reports of that resulting in poor performance, and we're working on some fixes.