[Linux][ChatGPT Desktop] Streaming assistant response is anchored mostly above visible conversation viewport

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

ChatGPT Desktop 26.825.51511 amd64.

What subscription do you have?

ChatGPT Plus.

What platform is your computer?

• Ubuntu 24.04.4 LTS (Noble Numbat)
• Kernel: 6.8.0-138-generic #138-Ubuntu SMP PREEMPT_DYNAMIC Fri Jul 31 22:41:49 UTC 2026
• Architecture: x86_64 / amd64
• Desktop: Ubuntu GNOME
• Session: X11

What issue are you seeing?

When an assistant response begins streaming in a long existing conversation, the response/conversation viewport is positioned such that most of the newly generated response is above the visible top boundary of the conversation area. Only approximately a footer-sized lower portion of the response remains visible at the top of the window.

The response itself continues rendering and scrolling normally. It appears that the text/conversation viewport or scroll anchor is vertically offset outside the visible area.

Other application UI elements—including sidebar, header, message composer, and general window layout—remain positioned correctly. The issue appears isolated to the conversation/streaming-response layer rather than the entire application viewport.

No conversation data loss has been observed.

What steps can reproduce the bug?

1. Open a long existing ChatGPT conversation in the Linux desktop application.
2. Submit a prompt that produces a response of several paragraphs.
3. Observe the response while it is streaming.
4. The conversation advances until most of the active assistant response is above the visible conversation area.
5. Only the final/lower portion of the streaming response remains visible near the top of the window despite substantial available space between it and the composer.
6. Manually scroll/reposition the conversation downward while generation is still occurring.
7. The response becomes visible again and subsequent streaming continues from the corrected position.

The issue was reproduced on August 31, 2026 at approximately 12:17 PM PDT.

What is the expected behavior?

The currently streaming assistant response should remain within the visible conversation viewport, or the viewport should track/reposition sufficiently to keep the active response readable.

Workaround

Manually repositioning/scrolling the conversation downward while the response is generating restores the response to the visible area. Once manually repositioned, subsequent streaming text continues from that corrected position.

Additional information

• Reproducible in ChatGPT Desktop 26.825.51511.
• Occurs under X11, so this reproduction is not Wayland-specific.
• Application chrome/layout remains correct.
• Content is not missing or failing to render; the apparent problem is the position/scroll anchor of the conversation or streaming-response viewport.
• The behavior resembles a nested scroll/viewport anchoring or vertical-offset regression.
• Similar cross-platform conversation-scroll/viewport reports include #41446 and #37872.

A screenshot was captured during the incorrect state before manual repositioning. It clearly shows the tail of the assistant response against the top edge, reaction controls beneath it, a large amount of available space below, and the composer correctly positioned at the bottom. The issue-creation connector used here cannot attach a local PNG directly, so the screenshot could not be embedded in this filing; it is available from the original ChatGPT conversation if engineering/support requests it.

The report names no files, tests, or entry points. Reproduce the issue in a long conversation while streaming, then trace the conversation viewport and scroll-anchor behavior; done means the active assistant response remains readable without manual repositioning.