Telnyx Voice AI Pricing: What an AI Phone Call Costs

Telnyx's Voice AI engine costs $0.05 a minute, covering speech-to-text, voice and turn-taking. Language model tokens and the phone line bill on top, and Telnyx's own estimate puts the total near $0.06 a minute. Dograh, the open-source AI voice agent platform we build, can use Telnyx as its carrier. You can self-host it or run it on Dograh Cloud.

Key Takeaways

• Telnyx bills the voice engine and LLM tokens separately from the phone line.
• Every call rounds up to 60 seconds, so short calls cost a full minute.
• Under Dograh, Telnyx bills only telephony and streaming, $0.0435 for 5 minutes.

A Telnyx AI call has more lines than its headline rate. We go through the voice engine, tokens, telephony and plans, then price one 5-minute call on Telnyx's own agent. The last part prices the same call with Telnyx as the carrier under Dograh, using your own model keys.

How Telnyx bills a voice AI call

Telnyx bills a voice AI call in three parts, and only one of them is the headline rate. Telnyx charges $0.05 a minute for the voice engine, bills the language model's tokens on top of that, and bills telephony separately by its own rules. All three land on one Telnyx invoice.

Rounding applies to the whole call. Telnyx's billing increment is 60 seconds per call, so a 15-second call bills as 60 seconds and a 61-second call as 2 minutes. Telnyx notes that its own estimates leave this rounding out, which means short calls will cost more than the estimate. Ringing time and failed attempts are not charged.

So any per-minute figure from Telnyx is a starting point. Your real rate depends on how long your calls run and which language model you choose.

The voice engine and LLM tokens

The voice engine is the part Telnyx runs for you. For $0.05 a minute it handles turn-taking, interruptions, tool calls and knowledge base retrieval, along with speech-to-text and text-to-speech hosted by Telnyx. The speech side includes Deepgram models and Telnyx's own, and the voices include Telnyx's own and others such as Rime, Inworld and Murf.

The language model is billed by token, not by the minute. On models Telnyx runs on its own GPUs, it says Kimi typically comes to about $0.006 a minute, assuming about 6,000 prompt tokens and 500 output tokens a minute with prompt caching. Anthropic and OpenAI models are billed at those providers' list rates on the same invoice. A larger model, or a long prompt that isn't cached, raises this line.

The engine rate and the token rates are the same on every plan. That matters when you read the plan discounts further down, because they don't touch either line.

Telephony, numbers and add-ons

Telephony is the third line, and it has its own rates. On the Telnyx Voice API, every minute carries a Voice API fee plus a SIP trunking fee, and audio streamed to an outside agent platform adds a streaming fee.

The SIP fees are "from" prices, and Telnyx says rates vary by destination and volume tier, so a call mix with many international or high-cost routes will sit above them.

Numbers are low and fall with volume. Telnyx's US numbers start at $1.00 a month for a local or toll-free number, then drop to $0.79, $0.50, $0.39 and finally $0.25 a number above 5,000. Adding text messaging to a number costs $0.10 a month more. An 800 toll-free number has its own terms, with $500 up front for the first 12 months, then $40 a month.

Open Source Alternative to Vapi / Retell

Self-hosted voice agent platform — no per-minute fees

dograh-hq/dograh

Star on GitHub

Pay as you go, Growth and Enterprise

Telnyx sells three plans, and for most teams the choice is about minimum spend. Pay as you go costs nothing up front, has no monthly minimum and includes 500 concurrent calls.

Growth asks for a $2,000 monthly minimum spend and gives 15% off voice, messaging and numbers, with automatic volume discounts and round-the-clock support. Enterprise starts at a $5,000 monthly minimum, with custom pricing on dedicated infrastructure and a private network connection.

Read the Growth discount closely before you commit. It applies to telephony, messaging and numbers, while the voice engine and token rates stay the same on every plan. On a typical AI call the engine is the largest line, so the 15% applies to the smaller part of the bill.

One 5-minute call on Telnyx's own agent

Here is one worked example, built from the lines in Telnyx's own all-in estimate: a 5-minute inbound call to a US local number on Telnyx's Voice AI agent. Because the lines are Telnyx's own, it shows what Telnyx itself expects such a call to cost.

That matches Telnyx's own figure of about $0.06 a minute, which it scales to about $13,700 a month at 230,000 minutes. The number adds $1.00 a month. A call that runs 4 minutes 10 seconds still bills 5 minutes, so the total holds for it too.

The engine is $0.25 of the $0.30, so most of this bill is set by one rate you can't change. The model line can move, depending on which model you pick.

The same call with Telnyx under Dograh

The other way to use Telnyx is as the carrier only, with an agent platform doing the work the voice engine does. You connect Telnyx to Dograh with an API Key, and Dograh creates the Call Control application with its inbound webhook when you save. Each call's audio then streams to the agent, so the Telnyx bill is telephony plus streaming.

Here is the same 5-minute inbound call, as a worked example.

Self-hosted Dograh has no per-minute platform fee, but servers, model keys and phone minutes still cost money. Dograh Cloud charges a 1¢-a-minute platform fee. That is the $0.05 above. The model line is yours to choose. You bring your own keys for any speech-to-text, language model and voice provider, or run open models on your own servers. Our breakdown of what an AI voice agent costs per minute prices that line by provider.

The two routes charge for different things, so compare like with like. The engine's $0.05 includes speech-to-text and the voice, while Dograh's line leaves those to providers you pick and pay directly. Choose Telnyx's bundle if one invoice matters most, and the Dograh route if you want to pick each model or run the platform yourself.

Transfers, clips and checking the bill

A few Dograh features change how this bill behaves over a month. Transfer Call works on Telnyx. The person is dialled while the caller hears hold music, and the two connect once someone answers. If nobody does, the agent gets a transfer_failed result and the call carries on. On Telnyx, that person is called as a new outbound leg and both join a conference, so a transfer adds that leg and conference minutes. Dograh does not use Telnyx's own transfer command, the one listed at $0.10.

The voice line is often the easiest model cost to cut. Pre-recorded audio plays a recording for the lines your agent says on every call and falls back to a generated voice for the rest, so fewer minutes reach your voice provider.

Each call also leaves a full trace, and Dograh's built-in QA can review calls after they end, so you can find the calls that run long or loop. Calls placed through Telnyx that are never answered still get a status and fire your webhooks, which keeps campaign counts honest. For how carriers and numbers fit together, see Telephony 101 for voice AI, and for the carrier choice itself, our comparison of Twilio vs Telnyx for AI voice agents.