Skip to content
FasterTokensGet an API key

About 0.6¢ a minute of conversation.

That is what DeepSeek V4.1 Flash on FasterTokens costs a typical voice agent. You pay per token from prepaid credit: no subscription, no minimum, no charge for a turn that fails.

New accounts start with $1 of credit, about 170 minutes of calls, and no card is needed to try it. Get an API key

Per minute of call

Your agent’s promptPer minutePer hour
A focused agentInstructions and a few tools: about 1,200 tokens0.6¢$0.36
A detailed agentLonger instructions and more tools: about 3,000 tokens0.9¢$0.54
A long playbookA knowledge base in the prompt: about 6,000 tokens1.5¢$0.90

Estimates for a 3-minute call with 5 model turns a minute, a 40-token spoken reply each turn, and one session ID for the call so the prompt is cached after the first turn. At the first row’s rate, $10 of credit covers about 29 hours of calls. Your own figure depends on how long your prompt is and how often your agent speaks; the dashboard shows what each day actually cost.

Per token

TokensPer million
InputThe messages and tools you send.$2.50
Cached inputInput that repeats what an earlier turn of the same session sent, such as your system prompt.$0.25
OutputThe reply, or the tool call.$6.00

You are charged once per turn, for the tokens of the answer you receive, however many providers we queried to get it. A turn that fails or times out before any output costs nothing, so falling back to another model is never paid for twice. Token counts are in every response’s usage.

How you pay

Top upA one-time card payment through Stripe. Credit is added within seconds of the payment clearing.$10 to $1,000
Auto refillOptional. When your balance falls below a level you choose, your saved card is charged and the credit added, so calls never stop for an empty balance. Off until you turn it on.e.g. $25 under $5
Starting creditAdded to every new account. No card needed.$1

When a balance reaches zero, requests are answered with HTTP 402 until credit is added. Running a platform or expecting high volume? Write to us.