Video explanation
Conversation packs accumulate along a bench. A breakpoint divides reused prefix from fresh work; an earlier prompt edit reduces the reusable region.
Reuse a repeated prefix
An identical prompt prefix can reuse prior processing, reducing repeated computation under provider-specific rules.
Where edits occur matters
Early prefix changes can affect later matches. Keeping stable content first can improve reuse.
Costs are not universal
Cache writes, reads, retention and pricing vary by platform. The video’s ratios are an example, not universal current pricing.