We run an AI interviewer that talks to candidates over phone calls. It started on a managed voice platform at roughly 10¢ a minute. The same interviews now run on a custom LiveKit pipeline at about 2.5¢ a minute, all in. For reference, Retell's pricing calculator defaults to 11¢ a minute and LiveKit
We run an AI interviewer that talks to candidates over phone calls. It started on a managed voice platform at roughly 10¢ a minute. The same interviews now run on a custom LiveKit pipeline at about 2.5¢ a minute, all in. For reference, Retell's pricing calculator defaults to 11¢ a minute and LiveKit Cloud's own calculator to about 4.8¢. Telnyx SIP trunk into LiveKit SIP Silero VAD plus LiveKit's turn detector (v1-mini) running on CPU xAI streaming STT, Gemini 2.5 Flash-Lite, xAI TTS Self-hosted agent workers and Redis for session state Per minute of an inbound call, at published list prices, assuming 600 characters of agent speech and about 3,000 input / 175 output tokens a minute: Line item Rate Per minute xAI TTS $15 per 1M characters $0.0090 LiveKit SIP fee (Ship plan) $0.004 per minute $0.0040 xAI STT (streaming) $0.20 per hour $0.0033 Telnyx inbound local $0.0032 per minute $0.0032 Gemini 2.5 Flash-Lite $0.10 in / $0.40 out per 1M tokens $0.0004 Turn detection (local CPU) none $0.0000 Workers, Redis, logs self-hosted ~$0.0051 Total ~$0.025 A few things I didn't expect: TTS is the biggest line. Almost a cent a minute, billed by character. Shorter agent turns save more than any model swap would. The LLM barely matters for cost. Four hundredths of a cent. Pick it for latency and quality, not price. Rate cards only get you to 2¢. The last half cent is workers, Redis and logging. No vendor lists that. On per-minute cost alone, building wins easily. Count engineer time and it gets closer. With about $600 a month of fixed infra and two engineer-days a month of upkeep (at $150k a year, 260 working days), a custom pipeline only beats an 11¢ vendor above roughly 20,600 minutes a month. Below that, stay on a managed platform and spend the time on the product. The full migration, with the numbers behind the 75% cut, is in our 2.5¢ voice AI case study. If you want to run your own volumes, there's a voice AI cost calculator on our site. Rates are from the xAI, Google, Telnyx and LiveKit pricing pages. Originally published on Axionry Engineering.
Key Takeaways #
- •We run an AI interviewer that talks to candidates over phone calls
- •This story was reported by Dev.to , covering developments in thedev space.
- •AI advancements continue to reshape industries — read the full article on Dev.to for complete coverage.
📖 Continue reading the full article: