Need Estimated Cost Breakdown Per Minute for Futur...
# supporta
Azhar
03/19/2026, 3:43 AMWe’re trying to understand the expected cost per minute for our existing setup and configuration, specifically for estimating future calls.
This is not about past usage or logs — we need a way to predict/estimate costs in advance based on our current setup.
We’re looking for a full per-minute cost breakdown, including:
Vapi base cost
LLM cost (based on model/token usage)
STT (speech-to-text) cost
TTS (text-to-speech) cost
Is there any way to get this kind of cost estimation per minute in the Vapi dashboard or via API?
Also, is there a way to estimate:
Token usage per minute for the configured model
Cost contribution of each component (LLM, STT, TTS) for a typical call
v
Vapi Support Bot
03/19/2026, 3:44 AMCurrently, the Vapi dashboard does not provide a built-in tool for estimating future per-minute costs by component. Most costs are direct pass-throughs from your chosen providers (STT, LLM, TTS), plus a small Vapi fee. For detailed, up-to-date pricing and to estimate costs for your setup, please review the [Vapi pricing page](https://docs.vapi.ai/pricing) or contact support for a tailored breakdown.
Tip: You can analyze historical costs in Boards using formulas like:
Copy code
Average Cost Per Minute = {{totalCost}} / ({{totalMinutes}} / 60)s
Shaunak
03/19/2026, 7:45 AMHi
Shaunak
03/19/2026, 7:46 AMThere's currently no built-in predictive cost estimator in Vapi - cost tracking is based on actual usage after calls complete. That said, here's how to estimate costs for your setup:
Component pricing Vapi passes through provider costs at-cost, so your per-minute estimate depends on your specific stack:
Vapi base fee — $0.05/min
STT - depends on your provider (e.g. Deepgram Nova 2 is ~$0.0059/min)
LLM - depends on model and token usage (e.g. GPT-4o is ~$2.50/1M input tokens)
TTS - depends on provider (e.g. ElevenLabs is ~$0.30/1K characters)
Shaunak
03/19/2026, 7:46 AMBest way to estimate your specific setup Run a few test calls and check the costs breakdown in the call object - it shows exact spend per component. Use that as your baseline and extrapolate by call volume and duration.