I have a little trick to share. While this isn’t a perfect fix, you can use chatgpt-4o to apply the changes o3 suggested and give you a “full code”.
Sometimes it misses, but this has been saving me o3 usage in some cases and it is slightly better in circunventing the “low output” of the reasoning models.
Something like “Please apply the proposed changes and give me a full updated code.”
Tweak according to what comes missing, as sometimes o3 will mess with something else that gpt-4o will not be able to handle right away.