Every conversation about semantic reuse reaches the same question: what is it worth on my fleet? Answer it with your own numbers.
The model is three outcomes: semantic reuse hits recover prefill computation from semantically similar context, exact-prefix hits reuse byte-identical prefixes, and misses recompute everything. The model always runs and generates a fresh answer; reuse removes the redundant compute, not the generation. The gap between the chart's two lines, exact-prefix alone versus with semantic reuse, is the number to pressure-test. The README inside the tool lists every formula and every simplification behind it.
If the gap looks meaningful on your traffic, bring your numbers to a conversation. Request a demo and we will pressure-test them together.