Routing cost calculator (free demo)
Estimated monthly cost
| Request class | Routed model | Routed estimate per month | Baseline estimate per month |
|---|---|---|---|
| Total (estimate) | |||
This is an estimate from public list prices and your own inputs. It is not a promise of savings. Your bill depends on your traffic, caching, discounts and quality needs.
* No cached-input price for this model was stated on the fetched pricing page. The calculator treats its cached tokens at the full input price, so the estimate for that class may run high.
Claude 4.7 and later models use a newer tokenizer that Anthropic says produces about 30% more tokens for the same text. Adjust token counts if you compare across providers.
Public list prices as fetched Nov 13, 2026. Check the provider's page before relying on them.
How the estimate works
For each request class, the calculator uses this formula:
monthly cost = requests × ( (input_tokens × (1 − cache_share) × input_price) + (input_tokens × cache_share × cached_input_price) + (output_tokens × output_price) ) / 1,000,000
In plain words:
- Input tokens that are not served from cache are charged at the model's input price.
- Input tokens served from cache are charged at the model's cached-input price. If no cached price is listed for a model, the calculator uses the full input price and marks that class with a footnote.
- Output tokens are charged at the model's output price.
- Prices are per 1 million tokens, so the sum is divided by 1,000,000 and then multiplied by the number of requests per month.
- The baseline uses the same classes, volumes, token sizes and cache shares, all priced at the baseline model.
The estimate leaves out cache write charges, cache storage fees, batch discounts, off-peak rates, regional data-residency uplifts, taxes and any negotiated discounts. Long-context and dated rates are only used if you pick those entries in the model list.
Price list used by the calculator
Public list prices as fetched Nov 13, 2026. Check the provider's page before relying on them. USD per 1 million tokens, standard (non-batch, non-priority) API tier. Entries marked "price not verified" are not selectable.
The full workbook: $19 one-time
The demo stops at 3 request classes. The workbook is a set of CSV files that import into Google Sheets with working formulas, plus a short how-to. It adds:
- Unlimited request classes. Add as many rows as your traffic has types.
- Monthly scenarios. Your routed and baseline totals at 0.5x, 1x, 2x and 5x volume, plus a 12-month table with a growth rate you set.
- A quality-gate column. For each class, write down how you will check that the cheaper lane is good enough before you switch.
- A sensitivity table. See how cost moves when output tokens grow or the cache share changes.
- A price sheet you can update. One row per model with source URL and fetch date, plus the exact VLOOKUP formulas to link it to your routing sheet.
Checkout is handled by Stripe. After payment you go straight to the download page.
About
I am Iggy Mwangi, an AI systems architect in Dallas–Fort Worth, with 25+ years of shipping software. From my site:
"I call my approach Compound Engineering: put the effort into the harness (context, checks, and fixed workflows) instead of waiting for a better model…"
Routing is part of that harness: decide which requests need which model, check quality, and look at the numbers before you change anything.
At one employer, cost-aware routing cut LLM spend 27% (Iggy Mwangi, VynixAI; method and baseline not published). That is my own past result. It is not an estimate of what you will see.
Related free tools and reading
- Context Window Optimizer and Rate Limit Planner (free browser tools)
- Token Window Economics (article)
- iggym.github.io