Claude token cost calculator
Free, instant, no signup. Enter input, output and cached tokens per request plus your monthly volume, and get per-request and monthly cost for every Claude tier — with cache hits priced at a tenth of input, which is where most agent spend actually goes.
Your workload
Rate: $1.20 in / $6.00 out · cache hit $0.1200 · cache write $1.20 — all per 1M tokens, read live from the Kunavo catalog.
Estimated cost
You save vs Anthropic list
$130/ month · ~60%
Caching accounts for $130 / month of that — 24,000 of your 30,000 input tokens billed at $0.1200 instead of $1.20 per 1M.
Estimate only. Reasoning models bill hidden reasoning tokens at the output rate, so a reasoning workload costs more than the output box suggests. Cache writes are billed separately at $1.20 per 1M and are not in this figure — they are a one-off per cache entry, not per request. Failed requests are never billed.
The cache row is the one that moves the number
A Claude bill for a chat product is close to the two-term formula: input plus output, and caching barely matters because every prompt is different. A Claude bill for anything agentic is not. A coding agent re-sends its system prompt, its tool definitions and a repo map on every single turn, so the same tens of thousands of tokens are paid for again and again — unless they are cached, at which point they cost a tenth as much.
That is why this calculator asks for cached tokens as a subset of the input rather than an extra field. Set it to zero and you get the same answer every other calculator gives. Set it to the fraction of your prompt that is stable across turns and you get the answer your invoice gives.
Two rates that differ from calling Anthropic directly
The first is the headline one: every Claude model here is billed under Anthropic's published rate, and the calculator prints both figures so the gap is checkable rather than asserted.
The second is quieter and only shows up on cache-heavy workloads. Anthropic charges a premium to write a cache entry, on top of the input rate. Kunavo does not pass that surcharge through — cache writes bill at the plain input rate. On a long-context agent that rebuilds its cache many times a day, that is a real line item, and it is the reason the footnote under the calculator states the cache-write rate explicitly instead of hiding it.
Both rates come from the catalog at render time, the same source the pricing table and the caching docs read. Nothing on this page is a hand-typed price.
Claude pricing, in more depth
Every tier, next to Anthropic's list
How to switch it on, and the rates
Why hit rates drop, and what to do
Claude, GPT and Gemini side by side
Subscription vs API
What happens when you hit them
Measured cost per task
Every model, vs official
How is Claude API cost calculated?
Claude bills per million tokens with input and output at different rates, so one request costs (input tokens ÷ 1,000,000 × input rate) + (output tokens ÷ 1,000,000 × output rate). Output runs roughly five times the input rate across the Claude tiers, which is why a short prompt with a long answer can cost more than a long prompt with a short one. Prompt caching adds a third term: tokens served from cache are billed at one tenth of the input rate, so any workload that re-sends the same context is cheaper than the two-term formula suggests.
What does a Claude cache hit cost?
A cache read is billed at 0.10× the model's input rate on Kunavo, matching Anthropic's own cache-read ratio. On a coding-agent turn that re-sends a 24,000-token repo map inside a 30,000-token prompt, that turns 80% of the input into tokens costing a tenth as much — which is the single largest lever on an agent bill, larger than picking a cheaper tier for most workloads.
Does Kunavo charge extra to write a Claude prompt cache?
No. Anthropic's own API prices a cache write above the plain input rate. Kunavo does not pass that surcharge through: cache-write tokens are billed at the model's ordinary input rate. That is set in one place in the code — the write ratio for the Anthropic protocol is 1 — and the cache-write figure printed under this calculator is read from it rather than typed in, so the page cannot quote a rate the API does not charge.
Which Claude model is cheapest for a coding agent?
Haiku is the cheapest per token, but the cheapest per finished task is usually a split rather than a single model: a strong tier for edits and Haiku for the background sub-tasks the client fires on its own. Claude Code already routes those sub-tasks to whatever is mapped to the Haiku tier, so setting that mapping is worth more than switching the main model. Enter both workloads separately above to compare them.
Does this calculator count tokens from pasted text?
No — this calculator prices token counts you supply, it does not tokenize text. Several pages on this results page do both; this one does not, on purpose, because a character-count estimate drifts badly by language and by tokenizer. Read real counts from the usage field your API responses already return, including the cache-read count, and put those numbers in.
Why is my real Claude bill higher than this estimate?
Three things a per-request estimate leaves out. Extended thinking bills reasoning tokens at the output rate even though they never appear in the response. Cache writes are billed once per cache entry and are not in the per-request figure above. And retries are billed — a request that consumed tokens and then failed downstream still counts. Agentic tools show the widest gap, because one user-visible task is many billed round trips.