The model is degraded incredibly in codex. It shows the last date the model is trained was Jun-2024? WTF???
Open 💬 4 comments Opened Jul 8, 2026 by YanhaoZhang
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
I do not know it is me or there are other people like me. As a pro member, I paied 200 USD per month for coding assistant. However, codex becomes more and more dump these days. Today, I simpliy check when did the model is trained, and it turns out the following result.
<img width="836" height="531" alt="Image" src="https://github.com/user-attachments/assets/50b61bcf-3aae-46a3-b30c-5a7153e2eca1" />
4 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
What's the point in posting this when you haven't even stated what model you're using/selected? Codex has over 4 models, each with multiple reasoning effort levels. Relying on the LLM output for the model/cutoff is not valid.
The model is gpt-5.5 xhigh. The most powerful model OpenAI has currently released. If you refer to the above three links, you can easily realize that the issue is not just me but happens among other members.
I tried the query with 5.5xhigh and got the same answer. But the model seems know what happened in 2025. For example, it can tell Pope Leo XIV, and some other events in 2025. It seems like it's a codex setting issue instead of model issue.