Hey guy, this call looks very overpriced : 2cf81a39-ec3c-4260-a9a7-db9a5e45dea6 $3,7 for 13min with gpt 4o and 11labs (28cts/min). This call with the same setting is 20cts/min : c6a70933-5374-4c00-94cd-d4fe7b9d3938 Both are higher than the annonced price (15cts/min)

@RealKhalass - c6a70933-5374-4c00-94cd-d4fe7b9d3938 This works out to roughly 18/min. The extra 3/min (beyond the 15 baseline) is coming from GPT-4 and STT usage. - 2cf81a39-ec3c-4260-a9a7-db9a5e45dea6 This is about 27/min. Again, it’s mostly the GPT-4 token usage that raises the total well above the baseline 15/min. The TTS also added 2–3/min more.

Thank, how to control this and be at 15/min ?

$0.15/min is just an approximate cost not the actual cot depends upon the call duration and calculate in real time and it involve other factors such as failed api calls etc.

I see thanks, but I still don't understand why between 2 calls with the same setting and prompt and without api calls or else, the price is so different

The difference is because of tokens used by LLM and TTS/STT usage.

And do you have advices to reduce the number of tokens used ?

Let me get the detailed breakdown for you.

Thank you

@RealKhalass ### Call 1 ID:

c6a70933-5374-4c00-94cd-d4fe7b9d3938

- LLM (GPT-4o): \$0.6652 - Prompt tokens: 247,989 - Completion tokens: 678 - Speech-to-Text (Deepgram nova-2): \$0.0573 - Telephony (VAPI): \$0.2636 - Duration: 5.27 minutes - Text-to-Speech (ElevenLabs): \$0 - Characters: 3,369 - Analysis: - Summary: \$0.0057 - Structured Data: \$0.0176 - Success Evaluation: \$0.0112 ### Call 2 ID:

2cf81a39-ec3c-4260-a9a7-db9a5e45dea6

- LLM (GPT-4o): \$2.4833 - Prompt tokens: 681,023 - Completion tokens: 1,611 - Speech-to-Text (Deepgram nova-2): \$0.1327 - Telephony (VAPI): \$0.6542 - Duration: 13.08 minutes - Text-to-Speech (ElevenLabs): \$0.3699 - Characters: 7,398 - Analysis: - Summary: \$0.0115 - Structured Data: \$0.0247 - Success Evaluation: \$0.0173

Thank you very much for your time, to conclude how can I reduce the number of token used by the LLM to align to the price you displayed ?

Price displayed, is approximate value not the actual value. Actual value is what you have seen for you calls so use your call costs as source of truth.

I see, because on RetellAI the price was close to 13cts, I don't understand why there is so much token used on vapi, I will check the settings, thank you

Tokens is based on call. Let me know if you require further clarifications.

Thank you, you already did a lot