Hello Vapi Support, We’re trying to understand why Azure STT (Vapi-managed) billing is higher than our measured user speech on a specific call. For call : 019a2619-9be9-7992-ac29-e76ee6536d27, the dashboard shows Azure STT cost = $0.25310806. In the STT selector the rate shown for Azure is $0.017/min (in pricing page). At that rate, the charge implies about 14.89 minutes of billable audio ($0.25310806 ÷ $0.017). Our own aggregation of role:"user" utterances for this call totals ≈1.4526 mins, so we expected roughly cost=0.0242103 h×$1.00/h=$0.0242103 If the underlying Azure plan is $1.00/hour. If the underlying Azure plan is closer to $2.00/hour ≈ $0.03333/min, then $0.25310806 would correspond to ~7.59 minutes, still above our measured user speech. That’s why this looks unexpected on our side. 1- Could you please confirm the exact billing model used for Azure (Vapi-managed) and help reconcile this call? 2- Do you bill both directions when “transcribe agent/both sides” is enabled? And if it’s disabled, can agent/TTS audio still be captured or billed in any path? 3- If Call recording is ON, does auto-transcribe recordings add minutes in addition to live streaming STT? 4- What rounding/minimums apply (per-utterance rounding, trailing-silence padding, minimum block per stream, etc.)? With many short turns, how much overhead should we expect? 5- Do reconnections or STT restarts start new minimum blocks? 6- Which exact Azure Speech SKU/region backs the “Azure — $0.017/min” option (pure pass-through of a negotiated rate, or any additional factor)? 7- Could you share a minute breakdown that sums to $0.25310806 (user live STT, agent live STT if any, recording-transcription minutes, rounding/blocks, restarts)? so we need to understand the math behind : "costs": [ { "cost": 0.25310806, "type": "transcriber", "minutes": 7.5968, "transcriber": { "provider": "azure" } } ]

Thank you for the detailed breakdown. To help us investigate further, could you please confirm whether "transcribe agent/both sides" was enabled for this call, and if so, whether you see any agent-side or recording transcription minutes included in the dashboard's cost breakdown? This will help clarify if the higher billed minutes are due to both sides being transcribed or additional recording transcription. Source: - [Vapi Changelog June 11, 2025](https://docs.vapi.ai/changelog/2025/6/11)

If you mean “transcribe agent/both sides,” is the recording enabled? Yes, we have "recordingEnabled": true for this call.

To clarify how Vapi tracks and displays transcriber costs, you can view the detailed cost breakdown for each call using the

costs

array, which includes per-component charges like

"type": "transcriber"

and the associated minutes. For example:

json
{
  "cost": 0.25310806,
  "type": "transcriber",
  "minutes": 7.5968,
  "transcriber": { "provider": "azure" }
}

This structure helps you see exactly how many minutes were billed and at what cost for each provider. For further reconciliation or minute-by-minute breakdown, please refer to the dashboard or API as described in the [Vapi changelog](https://docs.vapi.ai/changelog/2025/6/11). Source: - [Vapi Changelog June 11, 2025](https://docs.vapi.ai/changelog/2025/6/11)

@Vapi Support Bot @Vapi But how is the cost calculated in Azure STT? And how does 7.5968 minutes end up costing approximately 0.25310806? What is the math behind that?

Hey sorry for the delayed response. Could you please provide a more recent call id preferably made within the last 7 days?

No worries 🙂 You can check this call ID: 019a96ef-9f8b-7ffd-94af-d862788320b2 "cost": 0.03314861, "type": "transcriber", "minutes": 0.9993666666666667, "transcriber": { "provider": "azure" }

We are reviewing the logs and will get back to you with an update since this request will take longer to investigate

Could you please share the full details on this matter as soon as possible? We are in discussions with a high-potential prospect and need this information to move forward.

Thanks for sharing the call ID — we reviewed the cost details. For this call, Azure STT billed $0.03314861 for \~1 minute of audio because Vapi uses Azure’s enhanced transcription models rather than the basic STT tier. These enhanced models (e.g., improved accuracy / premium real-time features) cost roughly 2× the base Azure rate, which matches the minute-level cost you’re seeing. So the pricing in your example is consistent with Azure’s enhanced-model rate, not the standard $0.017/min tier.

Thank you for the update, Kyle. Just to clarify, if we want to use the standard $0.017/min tier as listed on the pricing page instead of the enhanced model, is there a specific flag or configuration option we can set to enforce that lower-tier STT model using Vapi managed passthrough model?

If you would like to use the lower tier, please use your Azure API key and link it to your VAPI account through the integrations page