01 / The Model
Cache tokens
Input the provider has already processed and kept warm, charged at a fraction of the usual rate. The reason a long conversation is not as expensive as it looks.
How you’ll hear it
“Why is the second request so much cheaper?”
“The provider cached the part that did not change.”
Avoid
Assuming every request costs full price. Repeated material can be much cheaper.
The longer version
When the start of a request matches a recent one, many providers reuse their earlier processing and charge a fraction for those tokens. Keeping standing instructions and reference material stable, and at the start, is how you benefit.