[macOS][ChatGPT Desktop] Selecting assistant response text causes severe slowdown on the latest build

Summary

On the latest ChatGPT Desktop app for macOS, selecting part of an assistant response with a mouse or trackpad drag causes the app — and subjectively the Mac as a whole — to become suddenly very sluggish.

I have fully updated the app using the normal updater, and the problem still reproduces on the latest version available to me as of 2026-08-16.

The important trigger is text selection inside a rendered assistant response. The slowdown begins when I start selecting response text.

This appears related to existing selected-text / renderer issues, but the symptom I am seeing on macOS is somewhat different:

• #29987 — macOS app restarts after repeated drag actions, including text selection.
• #36279 — Windows renderer reloads when drag-selecting assistant Markdown output.
• #21326 — older macOS regression where basic interactions including text selection caused severe UI lag and 100–170% renderer CPU usage. That issue was closed after one reporter saw an improvement, although another user later reported that high CPU/overheating remained.
• #38468 — recent macOS report of severe performance regression, high CPU/memory usage, and UI hangs on a newer build.

Environment

• Application: new unified ChatGPT Desktop / Codex app for macOS
• Platform: macOS
• App update status: fully updated to the latest version offered by the app updater as of 2026-08-16
• Input: mouse/trackpad text selection in assistant responses

I can provide the exact app build number and macOS/hardware information if needed.

Steps to reproduce

1. Open ChatGPT Desktop on macOS.
2. Open a conversation containing a normal multi-paragraph assistant response.
3. Move the pointer over the rendered assistant response.
4. Click/press and drag to select part of the response text.
5. Observe that the UI becomes severely sluggish during/after the selection operation.
6. Repeat with another section of assistant response text.

The problem is associated with selecting rendered response text rather than generation itself.

Expected behavior

Selecting text in an assistant response should behave like an ordinary text-selection operation and should not noticeably increase system load or make the UI / Mac sluggish.

Actual behavior

Starting a text selection inside an assistant response causes an abrupt and severe slowdown. The interaction becomes noticeably laggy and the computer feels much heavier to operate while the issue is occurring.

Updating the app to the latest available version did not resolve the problem.

Why this may be selection/renderer related

There are several existing reports around the same interaction path:

• #29987 specifically identifies repeated drag actions and Text selection: yes on macOS.
• #36279 reports assistant Markdown drag-selection immediately preceding an Electron renderer reload on Windows.
• #21326 reported a macOS renderer performance regression where selecting text and other basic UI interactions became slow while the renderer consumed very high CPU.

The current behavior may therefore involve the selected-text interaction / overlay / renderer path rather than a generic model, network, or generation-performance issue. This is only a hypothesis; I have not yet captured a profiler trace proving the root cause.

Diagnostics I can collect

If useful for maintainers, I can reproduce the problem while capturing:

• Activity Monitor CPU usage for ChatGPT / ChatGPT Helper (Renderer) / ChatGPT Helper (GPU) / WindowServer before and during text selection
• Memory usage before and during the event
• macOS sample/spindump of the renderer while the slowdown is occurring
• Screen recording showing the selection-triggered lag
• Exact ChatGPT Desktop build number and macOS version

I would be happy to add those measurements to this issue once captured.

Related issues

• #29987
• #36279
• #21326
• #38468