https://vapi.ai/ logo
Join Discord
Powered by
Being charged more than I have used.
# support
y
I am being charged around $0.26/min, when my assistant shows a cost of ~0.08$/min, the org id is b2ead67c-a0ed-4e90-b67a-0f9259ba3e8c, please help me understand this. The vapi support email chain wasn't helpful.
s
Hey @yourboygrover Pleaes share a call ID.
heads up that the cost in the assistant page is an estimated cost, while the real cost depends on the actual usage; some costs, like structured outputs, and other analyses, are not considered in estimate cost in the dashboard.
We calculate cost from actual usage reported by providers; there's no way to calculate the exact cost of a call beforehand, because all subprocessor usase change on context of the call. Not all 1-minute calls consume the same amount of tokens. If you provide me a call Id, I will check if something can be improved costwise.
y
019e74c2-0ada-7000-9eb9-d6c5872f3a0b, i don’t have structured outputs or any analytics setup
and if there are costs for those, they should be visible while setting up, no?
and any service should tell me the cost beforehand and while i’m developing it, i should have a clear price when building and choosing.
s
Can you share referencing on where it's showing 0.08$/min?
For this call, here's the cost breakdown- * Model (Claude Opus 4.5): $0.1899 * Vapi Platform: $0.0506 * Voice (ElevenLabs Turbo v2.5): $0.0462 * Transcription (Deepgram Nova-3): $0.0103 * Call Summary Analysis: $0.0018 * Knowledge Base: $0.0000 * Success Evaluation: $0.0000 Total Cost: $0.2988
Claude Opus is an expensive model; most of the cost is coming from that. Switching to a cheaper model will reduce your cost by a lot.
y
The opus model shows a cost of 0.00 cents, because its from bedrock.
s
Looks like a frontend bug, it's not showing the cost for Bedrock, it's showing correctly for normal Anthropic.
But model costs are deducted from credits; unless you are using your own API key, we charge on the basis of tokens consumed.
To get an estimate for your current assistant, please consider cost shown in normal Anthropic provider.
y
I wasn't aware that it would be the same as the Anthropic costs, I thought it was like that because of the fact it was locally hosted by Vapi maybe.
It doesn't say the cost of the model in the API request either, curl https://api.vapi.ai/assistant \ -H "Authorization: Bearer " So where was I supposed to confirm the fact that the cost isin't 0.08$/min.
s
> It doesn't say the cost of the model in the API request either, The cost is estimated; the actual cost depends on what you use. Each call includes a proper cost breakdown.
We only charge on the basis of what underlie subprocessors' usage, call API includes every breakdown, including how many tokens were consumed. The only fee Vapi charges is $0.05 per minute on the call.
y
yeah but according to your response there is a a front end bug, and it was showing $0/min, so it inaccurately estimated.
therefore giving me an inaccurate judgement.
s
The cost shown in the dashboard is an estimate; the real cost always varies and is purely based on your usage. I apologize for the confusion it might've caused, flagged this issue to our product team.
y
What about the cost that was used under the impression of the 0$/min cost for the model.
s
I understand the concern, and I agree that the frontend bug caused confusion by not displaying the Bedrock model cost correctly. That said, we're unable to issue a refund for usage that was successfully processed. Vapi charges based on the actual provider usage incurred during calls, and those costs were charged as the calls were completed.
y
But I wasn't able to see the costs beforehand. I just doesen't make sense because I am currently in production and making this model change can change customer reaction.
So when I was in development, I understood this is the perfect mix of agents and I went into production.
And if I had the cost visible that wouldn't be an issue today.
And the bug still isin't fixed on the frontend.
Do you think that is fair to me?
s
The team is very occupied right now, so this bug might take time. I have created a proper feedback report to the product team on the confusion it have caused.
It's ambiguous, and I agree that the confusion was caused on our side. However, I'm not the right person to make a final judgment on this. Could you please create a new support ticket on mail and add me in CC? (I DMed you my email.) I'll escalate it to the right person who can investigate further.
PreviousNext
Powered by