[app][macOS] ChatGPT uses 3–5 GB RAM and makes an 8 GB Mac unusable even with no project open

Open 💬 2 comments Opened Aug 4, 2026 by Nadeem-GER
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of the Codex App are you using (From “About Codex” dialog)?

ChatGPT 26.727.51351 (build 6119), bundled Codex app

What subscription do you have?

Paid ChatGPT account (exact tier not shown in app diagnostics)

What platform is your computer?

Darwin 25.5.0 arm64 arm; macOS 26.5.2 (25F84); MacBook Air M2 (Mac14,2); 8 GB unified memory

What issue are you seeing?

The new unified ChatGPT desktop app continuously grows to approximately 3–5 GB of memory and makes the entire Mac unresponsive.

This happens even after fully quitting ChatGPT, rebooting the Mac, and launching the app without any project or local folder connected. It is therefore not limited to indexing a large workspace.

Activity Monitor evidence from 4 Aug 2026:

  • ChatGPT main process: 3.33 GB
  • Codex (Renderer): 393.3 MB
  • second Codex (Renderer): 198.8 MB
  • Codex (Service): 147.6 MB
  • visible ChatGPT/Codex total: at least approximately 4.07 GB
  • physical memory: 8.00 GB
  • memory used: 7.02 GB
  • compressed memory: 2.26 GB
  • swap used: 4.01 GB
  • memory pressure: continuously yellow

The memory usage regularly rises to 4 GB and sometimes 5 GB. At that point the whole Mac stalls and normal work becomes impossible. Restarting temporarily resets the condition, but it returns.

<img width="993" height="601" alt="Image" src="https://github.com/user-attachments/assets/37ee0957-db18-4d8d-8c3c-6d1bd34f68d5" />

What steps can reproduce the bug?

  1. Fully quit ChatGPT with Command-Q.
  2. Restart the Mac.
  3. Launch ChatGPT 26.727.51351.
  4. Do not connect or open any project, repository, or local document folder.
  5. Leave the app open or use a normal Chat/Work session.
  6. Watch memory usage in Activity Monitor.
  7. ChatGPT grows to about 3–5 GB; memory pressure turns yellow, compression and swap increase, and the system becomes unresponsive.

The issue is reproducible after restart and without project files.

What is the expected behavior?

ChatGPT/Codex memory usage should remain bounded during normal use and while idle, including on a supported 8 GB Apple-silicon Mac. The app should not consume roughly half of physical memory by itself, force several gigabytes of swap, or make the operating system unresponsive. Background renderer and service processes should release unused memory.

Additional information

This appears related to the following open reports:

  • #11984: renderer becomes slow around 4 GB and unresponsive above 5 GB
  • #19333: high memory usage on an 8 GB M2 Mac, including while mostly idle
  • #33319: unified ChatGPT/Codex app workers accumulate memory and exhaust swap

The attached Activity Monitor screenshot shows the measured state. The macOS username visible in the screenshot is the reporter's own local account name. No project content or private documents are shown.

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 23 days ago

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

  • #36749

Powered by Codex Action

kevinmarty69 · 1 day ago

Additional reproduction on a newer macOS build:

  • Codex Desktop: 26.820.60940
  • Platform: macOS, Apple Silicon / arm64
  • Subscription: ChatGPT Pro
  • Observed memory: approximately 6 GB RAM while idle, with no active Codex command/tool execution running
  • Workload context: a long-lived Codex development thread with substantial history (many shell commands, Azure command outputs, diffs, test logs, file inspections, and context compactions)
  • The memory footprint remains high after the active task has completed instead of returning to a bounded idle baseline

This looks consistent with the memory-retention/leak behavior reported here, and may indicate that large/long-lived session history aggravates the issue. The process is idle from the user's perspective when the ~6 GB footprint is observed.

A full app quit/relaunch is currently the practical way to reclaim the memory.