Every Vapi pricing article says "starts at $0.05/minute." That is technically true and completely misleading. $0.05/minute is the platform fee only. Your actual cost per minute of a voice agent call is $0.15-0.40 once you add in the LLM, speech-to-text, text-to-speech, and telephony costs that Vapi charges on top. We have deployed 8 production Vapi agents. Here is what the bills actually look like.
How Vapi Pricing Actually Works
Vapi is a platform that orchestrates voice AI calls. It does not provide the LLM, TTS, or STT itself -- it connects them. You pay Vapi for orchestration plus each provider for their piece:
- Vapi platform fee: $0.05/minute
- LLM (your choice): $0.01-0.15/minute depending on model
- Speech-to-text (STT): $0.01-0.04/minute
- Text-to-speech (TTS): $0.02-0.10/minute
- Telephony (Twilio/Vonage): $0.01-0.02/minute
Real Cost per Minute: 3 Common Stacks
Budget Stack: ~$0.15/minute
- Vapi: $0.05
- LLM: budget OpenAI model ($0.02)
- STT: Deepgram Nova-3 streaming ($0.005)
- TTS: Deepgram Aura ($0.02)
- Telephony: Twilio ($0.015)
- Total: ~$0.115-0.15/minute
This is the cheapest production-ready stack. Deepgram Aura TTS is good but not as natural as ElevenLabs. A budget-tier model handles simple call flows well -- appointment booking, FAQ answering, basic lead qualification. But it stumbles on complex conversations or nuanced objection handling.
Standard Stack: ~$0.25/minute
- Vapi: $0.05
- LLM: GPT-5, Vapi's announced default ($0.08)
- STT: Deepgram Nova-3 streaming ($0.005)
- TTS: ElevenLabs Turbo v2 ($0.06)
- Telephony: Twilio ($0.015)
- Total: ~$0.22-0.28/minute
This is what we deploy for most clients. The voice quality upgrade from ElevenLabs is worth the extra cost -- callers cannot tell it is AI in about 70% of test calls. GPT-5 handles nuanced conversations and can follow complex call scripts with branching logic.
Premium Stack: ~$0.40/minute
- Vapi: $0.05
- LLM: Claude Sonnet 5 or GPT-5 with tools ($0.12)
- STT: Deepgram Nova-3 streaming ($0.005)
- TTS: ElevenLabs Multilingual v2 ($0.10)
- Telephony: Twilio ($0.015)
- Total: ~$0.35-0.42/minute
Reserved for high-value calls where quality is critical. Think sales calls for $10,000+ products, medical appointment triage, or financial services consultations. The premium voice and LLM make the AI nearly indistinguishable from a human agent.
Monthly Cost Projections
Here is what to budget based on call volume:
- 500 minutes/month (small business, ~25 calls/day × 1 min): $75-200/month
- 2,000 minutes/month (medium business, ~100 calls/day × 1 min): $300-800/month
- 10,000 minutes/month (high volume, call center replacement): $1,500-4,000/month
- 50,000 minutes/month (enterprise): $7,500-20,000/month
Compare that to a human call center agent at $15-25/hour ($2,400-4,000/month full-time). Even the premium Vapi stack at 2,000 minutes/month ($800) is 75% cheaper than one human agent handling the same volume.
| Provider | Vapi | Retell AI | Synthflow |
|---|---|---|---|
| Pricing Model | Orchestration Fee + Providers | All-in-one per minute | Enterprise contracts (from $30K/yr) |
| Real Efficiency | Best at Scale (10k+ min) | Best for Mid-Volume | Enterprise-Only Now |
| Flexibility | Maximum (any LLM/TTS) | Limited (internal models) | Moderate |
| PxlPeak Recommendation | Best ROI for Agencies | Best for Rapid Prototypes | Only at Enterprise Budget |
Vapi vs Retell AI vs Synthflow: Pricing Comparison
- Vapi: $0.05/min platform + separate provider costs. Most flexible, most complex pricing. Best for developers who want full control.
- Retell AI: ~$0.115/min all-in with GPT-4.1 (voice engine $0.055 + TTS $0.015 + US telephony $0.015 + LLM from $0.045/min). Simpler pricing, less model flexibility. Better for teams that want predictable bills.
- Synthflow: enterprise-only as of August 2026 contracts start at $30,000/year and the old $29-450/month self-serve tiers are no longer published. Only relevant if you want a fully managed no-code platform at enterprise budget.
See our detailed comparisons: Synthflow vs Vapi and Vapi vs Retell AI.
Hidden Costs and Gotchas
- Concurrency fees: Vapi charges extra for concurrent calls above your plan limit. If 5 calls come in simultaneously and your plan supports 2, the extra 3 either queue or fail.
- Twilio phone numbers: $1-2/month per phone number. If you need local numbers in multiple area codes, this adds up.
- Failed call charges: If the AI picks up and the caller hangs up after 3 seconds, you still pay for those 3 seconds across all providers. High-spam-traffic numbers can rack up costs.
- Development time: Vapi is developer-first. Building a production-ready voice agent takes 20-40 hours of development, testing, and prompt engineering. That is $2,000-4,000 in labor before you make your first call.
What We Actually Spend per Client
Our typical Vapi deployment for a small business:
- Stack: Standard (GPT-5 + Deepgram + ElevenLabs)
- Volume: 800-1,500 minutes/month
- Monthly cost: $200-400/month
- Phone numbers: 1-2 ($2-4/month)
- Total: $202-404/month
For most clients, the AI voice agent replaces a $3,000-4,000/month receptionist or after-hours answering service. The ROI typically hits positive within the first month.
We build custom Vapi voice agents for appointment booking, lead qualification, customer support, and after-hours answering. See our AI voice agent service or browse our Vapi implementation page.
Frequently Asked Questions
Is Vapi free to try?
Vapi offers $10 in free credits when you sign up. At the budget stack rate, that is about 65 minutes of calls -- enough to build and test a basic voice agent. No credit card required to start.
Why is Vapi cheaper than competitors?
It is not, once you add all the provider costs. Vapi quotes a low platform fee because you pay providers separately. All-inclusive platforms like Retell and Synthflow bundle those costs into a higher per-minute or subscription rate. The total is often similar.
Can I switch LLM/TTS providers on Vapi?
Yes, this is Vapi's biggest advantage. You can swap GPT-5 for Claude, switch from ElevenLabs to Deepgram Aura, or try different STT providers with a config change. No code changes needed. This flexibility lets you optimize cost and quality independently.