Fits the model, effort, and context to your coding agent's task. Fewer tokens, same quality. For Claude Code — and Codex CLI (experimental).