ChatGPT web: adding a follow-up during an active Pro response shows “Internal Server Error” (5.6 Sol Pro and 6 Pro)

Summary

On ChatGPT web in Chrome on macOS, sending a follow-up message while GPT-5.6 Sol Pro is still working can display Internal Server Error instead of incorporating the new instruction into the ongoing response.

I also encounter the same webpage behavior with 6 Pro. The detailed reproduction steps and repeated tests below were performed with GPT-5.6 Sol Pro.

I use this feature to refine a request while the answer is being generated. Waiting for the answer to finish would not meet this use case.

Environment

• Product: ChatGPT web at chatgpt.com
• Browser: Google Chrome
• Operating system: macOS
• Subscription: Pro
• Model selection: GPT-5.6 Sol, with the Pro capability setting (displayed as “5.6 Pro” in the composer)
• Also affected: 6 Pro, based on my additional observations
• Observed on September 7–8, 2026

Steps to reproduce

1. Start a new conversation and select GPT-5.6 Sol with Pro enabled.
2. Submit a request that takes some time to answer, such as asking it to search official sources and compare several features.
3. While the response is still running—for example, while search progress is visible and the composer still offers the Stop button—type an additional instruction.
4. Submit that instruction before the original response has finished.
5. Observe the response area.

An example prompt used in the reproduction, originally submitted in Chinese:

请检索 OpenAI 官方资料，详细比较 ChatGPT 网页的普通生成、后台长时间思考、生成中继续追问这三种情况，并逐项给出可核对的来源。请先检索再回答。

Follow-up sent while the response was still running:

请把重点改为生成尚未结束时继续追问，并在最终回答中说明这一功能。

Expected behavior

The ongoing response should incorporate the additional instruction and continue, without requiring the user to wait for completion or start over.

Actual behavior

• The follow-up appears in the conversation.
• The response area shows “Pro 思考中” and then Internal Server Error.
• The Stop button can remain visible after the error, leaving it unclear whether the request is still running or has failed.

Reproduction notes

• Ordinary responses can work normally in the same browser and account.
• The failure is not limited to one existing conversation: it reproduced in two separate new conversations on September 8.
• It can occur on the first follow-up sent during an active response. In an earlier test, one follow-up was accepted and a later follow-up during the same ongoing task failed.
• Clearing ChatGPT/OpenAI cookies and local site data, signing in again, and testing a new conversation did not eliminate the issue in these tests.

The observations above do not establish the internal cause or imply that every follow-up fails. They describe a repeatable failure of the in-progress follow-up workflow.

No files, tests, or entry points are named. Start by reproducing the follow-up workflow in ChatGPT web with Chrome on macOS while a Pro response is active, then trace the frontend handling of the in-progress follow-up and its error state. Done means the follow-up is incorporated without an Internal Server Error or a lingering Stop button.