Does anyone else feel like using GPT-5.5 for planning before handing things off to Codex 5.3 can get ridiculously token-heavy?
I’m on GPT Plus, and even for simple tasks, chatgpt GPT-5.5 thinking sometimes generates a super detailed plan with tests, edge cases, and tons of conditions. By the time I pass it to Codex 5.3 high, it feels like I’ve already burned through around 20% of my token budget.
I usually end up manually trimming the prompt, which kind of defeats the purpose.
How are you guys handling this?