GPT-6 Thinking Responses Stop Unexpectedly Mid-Generation Without Error

ChatGPT Bugs
1

Environment

  • Product: ChatGPT Web

  • Browser: Google Chrome 156 on Windows

  • Model: GPT-6 Thinking, particularly High / Extra High reasoning modes

  • Issue type: Unexpected response termination / incomplete generation

Bug Description

I am experiencing a recurring issue where GPT-6 unexpectedly terminates its responses before completing the requested task.

The problem occurs in two forms:

  1. Premature thinking termination: The model stops during its reasoning process without delivering a complete final answer.

  2. Mid-generation interruption: The model begins generating a normal response but suddenly stops midway through the output, sometimes before finishing a sentence or explanation.

In these cases, the interface may return to its normal idle state as though generation has finished successfully, despite the response being incomplete.

I did not manually stop the generation.

Steps to Reproduce

  1. Open ChatGPT and select GPT-6 Thinking with High or Extra High reasoning.

  2. Submit a complex question, multi-step task, or request requiring extended reasoning.

  3. Allow the model to process the request and begin responding.

  4. Observe that generation sometimes stops unexpectedly during reasoning or while displaying the final answer.

The issue is intermittent and does not necessarily occur on every request.

Expected Behavior

The model should complete its reasoning and deliver the full response, or display a clear error message if generation fails.

Actual Behavior

The response terminates prematurely without adequately explaining why. In some cases, the UI behaves as though the response has completed normally.

This makes it difficult to distinguish an actual completed response from an interrupted one.

Impact

This issue significantly affects the reliability of GPT-6 for academic research, programming, technical analysis, and other complex tasks.

Long-running reasoning becomes unreliable because users cannot be certain that the model will finish its work.

Request

Could the OpenAI engineering team investigate whether this is related to premature response termination, streaming interruptions, or incorrect completion-state handling?

It would also be helpful if interrupted responses displayed a clear error and provided a reliable way to resume generation rather than silently ending.

Similar symptoms have been reported by other users in the developer community.

Thank you for investigating this issue.