Same here. My experience so far comparing Visual Studio Code with the Codex Extension vs. Visual Studio Code with GitHub Copilot:
-
Codex is noticeably slower.
-
There’s no way to preview or track changes made by the agent (no Undo/Apply workflow like in Copilot), which makes it hard to see what actually changed between prompts.
→ Edited: This seems to be working now. -
I also ran into issues with
\r\ncharacters (as mentioned earlier) in Codex. -
I tested with gpt-5-codex-medium.
Overall, given these issues, it doesn’t feel usable in its current state. Would be great to see some improvements in the near future.
Now that we can also use the new Codex model within GitHub Copilot:
Has anyone tried Visual Studio Code + GitHub Copilot (with gpt-5-codex) and compared it directly to using Codex itself (same model)?strong text