Paste a prompt to see roughly how many tokens it uses, whether it fits the context window, and what it costs per call and per month - across every major provider rather than one. Arabic is counted separately, because tokenisers treat it very differently.
Paste a prompt, or open a file from your machine.
Detected mostly Latin script. This is an estimate from the character mix, not a real tokeniser - treat it as ±20%. The exact count is one call away; the snippets are at the bottom of this page.
What the monthly figure is multiplied by.
For a model that is not listed, a host charging something else for open weights, or a price you negotiated. Per million tokens.
30 of 30 models. Prices are per million tokens on each provider's own API.
$10 / $50cache read $0.25
$5 / $25cache read $0.5
$5 / $25cache read $0.5
$2 / $10cache read $0.2
$3 / $15cache read $0.3
$1 / $5cache read $0.1
$10 / $50cache read $1
$4 / $20cache read $0.4
$2 / $12cache read $0.2
$0.2 / $1.2cache read $0.02
$2.5 / $15cache read $0.25
$0.75 / $4.5cache read $0.075
$0.05 / $0.4cache read $0.005
$2 / $12cache read $0.2
$0.75 / $3.75cache read $0.075
$1.5 / $9cache read $0.15
$0.3 / $2.5cache read $0.03
$0.1 / $0.4cache read $0.01
$2 / $6cache read $0.5
$1.25 / $2.5cache read $0.2
$1.32 / $3.96cache read $0.044
$0.3 / $1.2cache read $0.006
$1.5 / $7.5no published cache rate
$0.15 / $0.6no published cache rate
$0.3 / $0.9no published cache rate
$1.04 / $1.04no published cache rate
$2 / $6no published cache rate
$0.15 / $0.47no published cache rate
$1.4 / $4.4no published cache rate
$3 / $15no published cache rate
A free endpoint that runs the model's own tokeniser and charges nothing for the call.
import anthropic
client = anthropic.Anthropic()
count = client.messages.count_tokens(
model="claude-opus-5",
messages=[{"role": "user", "content": text}],
)
print(count.input_tokens)A real tokeniser means shipping a vocabulary of tens of thousands of entries - and the wrong one for whichever model you use, since every provider tokenises differently. This page estimates from the character mix instead, and tells you its margin.
Most tokeniser vocabularies were built mainly on English text, so Arabic words break into several pieces and diacritics often cost a token each. The same paragraph can cost fifty per cent more in Arabic than in English - this page accounts for that.
Each provider's own published page, read on 2026-09-20 and linked above. Tiered rates, promotional rates and off-peak discounts are noted on the row they apply to. Cloud marketplaces such as Bedrock and Vertex are priced separately and are not modelled here.