dual-salmonD
VAPI•3mo ago
dual-salmon

models on vapi have a base price + a price per token based on sys prompt?

hey.. as stated

how do I know?

with such system prompt :

Tokens: 3798
Characters: 16235

with claude sonnet 4.5

I see price : 0.14761 $ /min (only for model)

with NO system prompt, Sonnet 4.5 too

I see Cost (USD) 0.0015 $ /min

So, I assume there is a base price, in 4.5 sonnet case : 0.0015, to which we add an estimation per minute of tokens used? I guess?

using price per 1M / token displayed here : https://vapi.ai/pricing

am I correct?

can we know more about the estimation?

would it be possible to have the formula instead of reverse eengineering them all?

would help a lot, as I'm building my pricing on top of vapi's

thanks!

here is what claude told me

is it right?


Token cost calculation:
Base cost (no system prompt):

0.0015 USD/min

With your system prompt:

0.14761 USD/min

Difference:

0.14761 - 0.0015 = 0.14611 USD/min from system prompt

Your system prompt tokens:

3798 tokens

Price per 1M input tokens (Sonnet 4.5):

$3/1M input = $0.000003 per token

Formula they're using:
cost_per_min = (tokens × price_per_token × messages_per_min)
Checking math:
0.14611 / 60 seconds = 0.002435 per second
0.002435 / 0.000003 = 811.67 tokens per second
811.67 / 3798 = 0.214 messages per second
= 12.8 messages per minute
Their estimation formula:
system_prompt_tokens × $0.000003 × ~13 messages/min ≈ cost/min
3798 × 0.000003 × 13 = 0.148 USD/min ✓
They assume ~13 message exchanges per minute to calculate the per-minute cost impact of system prompts.
CleanShot_2025-10-12_at_19.15.402x.png
Vapi
Vapi - Build Advanced Voice AI Agents
Build, test, and deploy advanced voice AI agents in minutes with Vapi. The platform for developers creating conversational voice AI.
Was this page helpful?