ai model routing calculator
Most workloads are a mix: many easy requests, some ordinary ones, a few hard ones. Sending them all to the strongest model is simple and expensive. Put in your mix and the calculator shows what a router saves after paying for the routing call and for the requests that fail on a cheap model and have to run again.
routing saves $1,344 a month (69%)
| tier | model | requests | retried | cost |
|---|---|---|---|---|
| simple | Haiku 5.5 | 60,000 | 4,800 | $76 |
| standard | Sonnet 5.5 | 30,000 | 1,500 | $321 |
| hard | Opus 5.5 | 10,000 | 0 | $194 |
| router | Haiku 5.5 | 100,000 | – | $9 |
Routing still pays even if every simple request is retried one tier up. A retried request is paid twice. List prices checked 2026-11-10; cache writes are left out.
questions
how does the router decide?
Usually a small, fast model (or plain rules) reads the request, or its first part, and labels it simple, standard or hard. The cost of that call is included. Rules cost nothing per request, so set the router's tokens to zero if you route with code.
what is the retry rate?
The share of requests a cheaper model gets wrong, or flags as low confidence, that you then rerun one tier up. You pay for both attempts. The calculator shows the retry rate at which routing stops paying, so you know what to measure in a pilot.
does it include prompt caching?
Cache reads, yes: set the share of input served from cache. Cache writes are left out because they depend on traffic patterns; the token cost estimator includes them.
which prices does it use?
Claude API list prices per million tokens, the same table as the token cost estimator, with the date they were checked. Haiku prompts over 100K tokens bill at its long-context rate.
more free tools
- ai roi & token cost estimatorMonthly Claude API cost, cost per task and ROI for a workload
- eu ai act compliance auditorRisk tier, obligations and dates for one AI system
- agent workflow fit scorerIs this process a good fit for AI agents, and what shape?
- ai build vs buy calculatorThree-year cost: SaaS tool, custom agents, or keep it manual
- ai accuracy proof calculatorHow many test cases prove a vendor's accuracy claim?
- human review staffing calculatorHow many reviewers an AI workflow's escalations need
- ai vendor scorecardWeighted comparison of AI vendors, with must-haves and a swing check
- gdpr dpia screener for aiDoes this AI use of personal data need a DPIA?
Want the numbers checked against your own process? Book a 20-minute call.