GPT-5.1 model returns prematurely with only ~20-30% of tasks completed

What version of Codex is running?

0.63.0

What subscription do you have?

Pro

Which model were you using?

gpt-5.1-high

What platform is your computer?

_No response_

What issue are you seeing?

The GPT-5.1 model in Codex has become extremely lazy. For complex tasks that should run for several minutes and complete all requested jobs before replying, the model now stops after about 1‑2 minutes and returns partial results – often only 20‑30% of the tasks are done. The remaining tasks are not performed unless the user repeatedly enters "continue" to prompt the model. There are no explicit error messages; the session simply terminates prematurely with incomplete work. Many users on social media have reported the same issue.

What steps can reproduce the bug?

When using the GPT-5.1 model inside Codex, start a complex request that normally requires multi‑step reasoning and several minutes to complete. In the past, Codex would think for 4–5 minutes (or longer) and return only when all jobs were finished. Now, after about 1–2 minutes the model stops and responds even though only 20–30% of the requested work has been completed. The rest of the tasks are not handled and the user has to repeatedly send “continue” commands to get the model to finish. This behaviour started around Nov 23–24 2025.

What is the expected behavior?

The model should continue processing until all requested tasks are completed, as it did previously. It should not terminate early after only 1–2 minutes with partial results. The user should not need to repeatedly request "continue" for the model to finish the job.

Additional information

Many users on social media (e.g., Twitter) have reported similar laziness issues with GPT-5.1. It might be related to a recent update or system prompt. Please investigate why the model behaviour changed and restore its full processing behaviour.