Price the workload before you build it

Two things this deliberately leaves out

Caching. Repeated context is billed at a fraction of the price and identical repeats cost nothing at all, so a real bill for agent traffic usually lands under these figures - but by how much depends on your traffic, and a calculator that guessed would be flattering rather than useful.

Routing. These are single-model figures. Sending "auto" spreads a mixed workload across models by what each request actually needs, which is the whole argument for a router - and the saving is measured per request rather than estimated here.

The one number not shown above is our fee, because it is not charged per request: it applies once when you buy credit. The pricing page has the arithmetic.

The same answer, from your code

For anything programmatic - pricing a job before running it, or showing a customer what their request will cost - there is an endpoint that uses the biller's own code rather than a copy of it.

curl https://finalrouter.com/api/v1/cost/estimate \
  -H "Authorization: Bearer $FINAL_ROUTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5",
    "messages": [{ "role": "user", "content": "..." }],
    "max_tokens": 800
  }'

# Free, works on an empty balance, and runs the same
# arithmetic the biller runs - so the estimate cannot
# drift away from the invoice.