Bring your own keys / one router in front of every model
Most requests don't need
a frontier model.
turnfare reads what you are actually asking for, then sends it to the cheapest model that can do it.
How it works
Three steps. Only the first one guesses.
- 01
Read the request
Before anything is sent to a provider, turnfare reads what the request actually is - a quick edit, a data extraction, a long piece of reasoning - and how much work it will take.
- 02
Price every option
Each model you have connected is priced for that specific job. Not just the list price: also what it costs when a cheap model cannot finish and you end up paying twice. The cheapest one that will actually do the work wins.
- 03
Show the receipt
Every request comes back with the model it chose, what it cost, what it would have cost, and the reason. If you disagree with a route, you can override it.
Where it stands
Every number here comes from
the model that just answered you.
41%
cheaper on a mixed workload
Measured end to end against sending every request to claude-fable-5, across a set of prompts the router has never seen. An average, not a best case.
65%
of requests read correctly
Across six kinds of work, on 150 prompts written by hand and never used for training. When it is unsure it routes up, because a wrong answer costs more than a slightly dearer one.