Token efficiency is where things get interesting. GPT-5.5 uses 72% fewer output tokens than Opus 4.7 on the same coding tasks. Fewer tokens means lower cost per task, even though GPT-5.5 costs $30/M output vs Claude's $25/M. But Claude's time-to-first-token is roughly 0.5s vs

→ View original post on X — @godofprompt