GitHub Copilot has an "auto" mode that automatically selects a model.
Analyzing the models I used for a month in June, it turns out I used GPT 5.4 and 5.3 codex the most among those with "Auto" in front. Those without "Auto" are models I manually selected. Thinking about it, I use Claude haiku quite a bit, so it seems like it might be able to produce more satisfactory results for things like sonnets or operas that I do. The results are already satisfactory.
I used to think Claude was amazing and only used that, but since I switched to auto mode to save costs, it's interesting to see that codex works better for my work.
By the way, it resets at 12 o'clock and I still have about 5,000 credits left. ㅎㄷㄷ What a waste.
