🧵 Comparison – Actual Call Content (Bot + User Messages) 🔴 Vapi Key Call (a847c3e6-...)

json

{
  "role": "bot",
  "message": "Guten Tag, mein Name ist Mia vom Kundenservice der BT Group. Wie kann ich Ihnen helfen?"
},
{
  "role": "user",
  "message": "Tell me, how are you?"
},
{
  "role": "bot",
  "message": "Thank you for asking. I'm here and ready to support you. How can I assist you today?"
}

✅ Own Key Call (4f004ae9-...)

json

{
  "role": "bot",
  "message": "Guten Tag, mein Name ist Mia vom Kunden der BT Group. Wie kann ich Ihnen helfen?"
},
{
  "role": "user",
  "message": "Hello, Mia. How are you?"
},
{
  "role": "bot",
  "message": "Hello. I'm doing well. Thank you. How can I help you today?"
}

🧠 These are semantically identical conversations – same length, same system prompt, same structure. 🚨 This Indicates a Severe Accounting Problem • Token usage with your key almost doubles even for tiny calls. • With longer calls, the token cost difference scales up drastically, indicating something is deeply off. • According to your own logs, 21,586 prompt tokens were used for a 15-second call with no tools, no long chat history. • With our own OpenAI key, the exact same content only used 10,811 prompt tokens – and was free under OpenAI’s token quota. 🧭 Our ask Please investigate: • How tokens are being calculated internally • Whether prior instructions or tool metadata are accumulating or duplicated • Whether a full LLM call payload (visible to us) could be exposed for debugging We’re fully prepared to assist with debugging this issue – but at the current rate, token usage (and therefore pricing) is non-viable for production-grade deployments.

### LLM Prices are also wrong! Also the LLM Prices are wrong! From your pricing page: Large Language Model (LLM) Advanced language processing and generation OpenAI GPT-4.1 - $0.06 / 1M tokens GPT-4.1 Mini - $0.01 / 1M tokens GPT-4.1 Nano - $0.01 / 1M tokens GPT-4.5 Preview - $2.12 / 1M tokens GPT-4o Mini - $0.01 / 1M tokens GPT-4o - $0.07 / 1M tokens ChatGPT-4o (Latest) - $0.14 / 1M tokens ## Example CallID: a847c3e6-cfbc-497f-bf62-21e397a32d95 From Statistic: llm $0.04 ($0.15 / min) ‣ 21586 prompt tokens ‣ 42 completion tokens > This is not 0.06 / 1m token, this is 1.85 / 1m token! > This are 3080% By your price list it would be: 0.06 / 1.000.000 * 21586 = 0.00129516 $ In normal Calls it's a very huge discrepance! CallID: 2afe8baf-3eb4-4b8f-b47f-46a1e95de288 llm $0.46 ($0.19 / min) ‣ 227522 prompt tokens ‣ 384 completion tokens > This is not 0.06 / 1m token, this is 2.02 / 1m token! > This are 3366% By your price list it would be: 0.06 / 1.000.000 * 21586 = 0.01365132