Windows Desktop 26.825.51511 – Long chats can lock without warning, Work long-paste hangs, and collapsed messages cannot be expanded

[TeknikerNL]

I want to report three separate regressions/issues that are seriously affecting my use of the ChatGPT Windows Desktop application.

Environment

• ChatGPT Desktop – Powered by Codex & OWL
• Windows
• Version 26.825.51511
• Released 30 Aug 2026
• ChatGPT Pro

These are three separate issues, although all of them appeared during the recent Desktop update cycle and together they are significantly disrupting long-form professional use.

1. Regular ChatGPT conversations can suddenly become locked without useful advance warning

This issue happens in normal ChatGPT conversations, not Work.

I use some ChatGPT conversations for long-running technical projects. After a conversation becomes sufficiently long, the application can suddenly show an error / conversation-length-related message and the thread becomes effectively unusable.

The biggest problem is that there is no useful advance warning.

By the time the limit/error appears, I may no longer be able to ask ChatGPT to create a final handoff or project-state summary for a new conversation.

This makes the failure particularly disruptive because the conversation may contain a large amount of accumulated project context, architectural decisions, test results, rejected approaches and pending work, yet the user is given no opportunity to safely close out the thread before it becomes unusable.

A conversation length limit itself is understandable.

The problem is the absence of graceful recovery.

Expected behavior:

• Warn the user well before the conversation reaches a hard limit.
• Preserve enough capacity for a final handoff/checkpoint.
• Offer a reliable “Continue in a new conversation” function.
• Transfer or summarize the authoritative state of the old conversation before it becomes unusable.
• Never allow a long-running conversation to become effectively locked without giving the user a chance to preserve continuity.

I have now experienced this more than once, including again today.

2. Long pasted prompts in Work can enter an indefinite processing loop

This is a different problem and specifically affects Work.

Previously, when I pasted a very long implementation prompt directly into Work, the application could automatically handle the oversized pasted content as an attachment and continue processing it.

That workflow was extremely useful for professional Codex/Work usage.

After the recent Desktop updates, it no longer works reliably.

Current behavior on 26.825.51511:

1. I copy a long structured prompt.
2. I paste it directly into Work.
3. The send button becomes visually greyed out, although it can still be clicked.
4. If I click it, Work enters a processing/loading loop.
5. It remains stuck indefinitely and the task never starts.

However, the exact same prompt works normally if I manually put it into a .txt file and attach that file to Work.

So the prompt content itself is clearly not the problem.

The current workaround is:

1. Open a text editor.
2. Create a .txt file.
3. Paste the Work prompt into it.
4. Save the file.
5. Manually attach it to Work.
6. Then Work processes it normally.

This strongly suggests that the regression is in the direct long-paste / automatic pasted-text attachment workflow rather than in Work’s ability to process the prompt itself.

Previously this conversion happened transparently. I could paste a long instruction and continue working without manually creating files.

Expected behavior:

• If pasted content exceeds the normal composer limit, automatically convert it into a usable text attachment as before.
• Preserve the complete pasted content.
• Allow Work to process it immediately.
• Never leave the user with a grey send button followed by an indefinite loading loop.

Long structured implementation prompts are normal in professional Codex/Work usage and should be treated as a first-class workflow rather than an edge case.

3. Long user messages are collapsed into black cards that cannot be expanded or edited during an active conversation

This third issue affects both normal ChatGPT conversations and Work.

Previously, when I sent a long prompt, long code block or other large message, ChatGPT displayed my message in a shortened/collapsed form.

That was perfectly fine because there was an expansion control underneath it. Clicking it immediately opened the complete message, allowing me to read everything I had submitted.

That behavior is now broken.

Current behavior on 26.825.51511:

1. I send a long prompt, code block or other large message.
2. My message appears inside a collapsed black-background card.
3. Only part of my submitted message is visible.
4. An ellipsis / expansion control (...) appears underneath the shortened content.
5. Clicking the control does absolutely nothing.
6. I therefore cannot immediately view the complete text of my own submitted message.
7. I also cannot reliably edit the complete message while it remains in this state.

This happens in both regular ChatGPT conversations and Work.

There is also a very unusual timing-related behavior.

If I leave the conversation alone for some time — sometimes approximately one or two hours — and return later, the exact same message may suddenly become expandable and editable again.

The underlying message therefore does not appear to be lost.

The problem appears to be related to the immediate UI/rendering/state of the collapsed message during an active conversation.

This is extremely inconvenient during technical work.

After submitting a long instruction, I often need to immediately reopen it to:

• verify exactly what I instructed;
• inspect a piece of code;
• copy part of my own prompt;
• check a technical parameter;
• or edit/correct something I just submitted.

At the moment I cannot reliably do that.

Waiting one or two hours for my own message to become readable or editable again is obviously not a usable workaround.

Expected behavior:

• Long messages may be collapsed for readability.
• Clicking the ellipsis / expansion control should immediately display the complete message.
• The complete submitted message should always remain accessible.
• Edit functionality should remain available immediately after submission.
• The behavior should work consistently in both ChatGPT and Work.
• A user should never have to wait for some later UI/session refresh before their own submitted message becomes readable.

Overall impact

Individually, these may appear to be separate UI or workflow bugs.

Together, however, they significantly degrade professional use of ChatGPT Desktop:

• long ChatGPT conversations can suddenly become unrecoverable without useful advance warning;
• long Work prompts that previously worked through automatic pasted-text handling can now enter an indefinite loop;
• and long messages that were successfully submitted may not even be immediately readable or editable.

These are not requests for new functionality.

These are workflows that previously worked and have regressed after updates.

This is especially frustrating because updates are expected to make an application more reliable and functional, not require additional manual work to accomplish tasks that previously worked correctly.

For users relying on ChatGPT Pro, Codex and Work for real technical projects, continuity and stability are extremely important.

I have also reported or confirmed related issues on GitHub.

All three problems described above are still present for me on:

ChatGPT Desktop – Powered by Codex & OWL
Windows
Version 26.825.51511
Released 30 Aug 2026

I would strongly encourage the team to prioritize regression testing for:

• long-running conversations;
• conversation-limit recovery and handoff;
• long pasted prompts;
• automatic pasted-text attachment handling;
• collapsed message expansion;
• message editing;
• and continuity between active Work/ChatGPT sessions.

New features are welcome, but preserving existing workflows that already worked reliably should come first.