Where exactly are they behind?

They’re about even in general, but for me OpenAI is slightly or significantly ahead in the areas I care about the most. E.g. claude code is a backend slop cannon if you don’t tell codex/gemini to review the outputs.