Cache savings
Prompt Caching Calculator
Prompt caching can change the economics of coding-agent work because repository summaries, system instructions, and long context blocks are often reused across turns. The benefit depends on cache write cost, cached input cost, and how much context repeats.
Use the main calculator to compare a no-cache estimate against cached input and cache write assumptions.
When caching matters
- Repository-wide refactors where the same files or summaries are reused.
- Long review loops where the agent keeps revisiting prior context.
- Multi-step feature work with repeated planning and verification.
- Team workflows with a shared system prompt or coding standard.
When caching matters less
Small one-shot prompts and short single-file edits may not repeat enough context to produce meaningful savings. In those cases, output tokens and agent turns usually dominate.