Methodology — where every number on this site comes from

Last updated 2 October 2026

Every rate, formula and assumption used on this site, separated into what the vendor publishes and what we estimate.

This page exists so that no number on the site has to be taken on trust. It separates what the vendor publishes from what we assume, gives the formula the calculator actually runs, and lists what the tool deliberately does not model.

The two kinds of number on this site

Published rates. Token prices, cache read and cache write prices, and context limits are read from the vendor's own pricing documentation. These are facts with a date attached, not opinions.

Planning assumptions. Anything about how much work a given effort level causes is an assumption, because no vendor publishes a fixed token multiplier per level. Anthropic describes effort as a behavioural signal rather than a strict token budget: the same level produces different amounts of work depending on the prompt and the task, and the levels are recalibrated between model generations.

Every multiplier in the calculator is editable for exactly this reason, and the UI labels it as your assumption the moment you touch it.

Rate sources

Rates are stored in a single data file that every page reads from. That file carries the verification date, and the footer of every page renders it, so a stale figure is visible rather than silent.

The formula

The calculator computes cost per call as:

cost = input_tokens        x input_rate
     + cache_write_tokens  x cache_write_rate
     + cache_read_tokens   x cache_read_rate
     + (visible_output + thinking_tokens + tool_tokens) x output_rate

Monthly figures multiply the per-call result by the call volume you enter. Nothing is amortised, discounted or averaged across a billing month.

The point worth repeating: thinking tokens sit on the output line. They are generated by the model and they count against max_tokens, so they are billed at the output rate — typically several times the input rate.

Effort multipliers

These are a starting point for routing decisions, not a vendor specification. They exist because you need some number to compare options with, and a documented assumption you can tune beats an undocumented one.

If you have real traces, measure your own ratios and replace them. The arithmetic on this site does not change when you do — only the column values move.

What the calculator does not model

• Retries and failures. A request that fails and is re-sent costs twice. The calculator counts successful calls only.
• Batch and asynchronous discounts. Where a vendor offers a reduced rate for non-interactive traffic, that is not applied. Treat the output as a conservative upper bound.
• Regional or data-residency surcharges. Not applied.
• Per-model default effort. The calculator uses the effort you select, not the level the model would have used if you had sent nothing. Those differ — most models default to high, and Opus 5.5 defaults to medium — and the difference is worth checking before you assume you are running at a low level.
• Latency. The calculator is about money. Higher effort also means slower responses, which is a separate trade-off.

Model defaults and unsupported combinations

Setting effort to a model's default is identical to omitting the parameter entirely, so "I never set it" does not mean "it is off".

Effort levels are also per-model capabilities. A model can reject a combination it does not support — for example, a request that disables thinking entirely while asking for xhigh or max can return an error rather than being silently downgraded. The calculator only offers levels a model accepts, but you should still validate the level you send against the model you send it to, and keep it in configuration rather than hard-coded.

Corrections

If a rate on this site is out of date, or a table disagrees with the calculator, tell us. Corrections are applied to the page itself and the "updated" date changes, so there is one current version rather than an accumulating errata list. See the contact page.

Independence

ClaudeEffort is not affiliated with, endorsed by or sponsored by Anthropic or OpenAI. Product names and prices referenced on this site belong to their respective owners. The site is funded by advertising, and no advertiser has any influence over the recommendations the tool makes.