Ai

I asked several AI assistants to compare and score each other. Here is the prompt I used, the results, and a short conclusion.

The /compact command summarizes a long coding session into a smaller context. It can save tokens and preserve direction, but it can also lose details that matter.

Claude Code can appear to burn more tokens than Codex because of how much context it carries, how often it summarizes, and how aggressively agentic coding tools inspect files and tool output.