Where ModelPilot cuts your Claude bill
We route each request to the cheapest model that's good enough — and prove it on your own traffic. Here are the work types we route down, and the teams it saves for. Your hard reasoning stays on the top model, untouched.
By work type
| Work type | Routes to | Typically routed | Cut vs Opus | Cut vs Sonnet |
|---|---|---|---|---|
| Short Q&A / lookups | Haiku | 100% | ~80% | ~67% |
| Simple code / SQL | Haiku | 100% | ~80% | ~67% |
| Rewrite / reformat | Haiku | 91% | ~73% | ~61% |
| Data / field extraction | Haiku | 89% | ~71% | ~59% |
| Translation | Haiku | 87% | ~70% | ~58% |
| Classification / triage | Haiku | 85% | ~68% | ~57% |
| Summaries (short) | Haiku | 75% | ~60% | ~50% |
| Summaries (long / dense) | Sonnet | 66% | ~26% | — |
| Conversation / advice | Sonnet | 40% | ~16% | — |
"Cut vs Sonnet" is lower because a Sonnet-baseline workload has less headroom — and the Sonnet-target rows show "—" (you're already at that tier).
Who on your team this saves for
| Team / role | The work we cut costs on |
|---|---|
| Customer support / CX | triage, ticket & thread summaries, draft replies, FAQ answers, translation |
| Operations / back-office | document & form extraction, routing & tagging |
| Legal / claims / clinical / compliance ops | contract & record extraction, document summaries, classification |
| Sales / marketing ops | first-draft emails & copy, lead enrichment, summaries, translation |
| Analysts & engineers (routine work) | simple SQL & queries, data extraction, snippets, lookups |
Hard reasoning — complex coding, debugging, math, agents, open-ended analysis, creative long-form — stays on the top model (quality protected, ~no savings). The savings come from the high-volume routine work above.
Figures are illustrative ranges at public list prices; your savings depend on your
traffic mix and current model. We measure the real number on your own traffic with a held-out
control arm — and you pay only a share of what we actually save (no savings, no bill).