Telnyx Voice AI $0.05 a minute with STT and TTS bundled

Telnyx prices its Voice AI engine at $0.05 per minute with orchestration, speech-to-text and text-to-speech in one rate, and says a production agent lands near $0.06 a minute once an LLM and telephony are added. Sume prices the same two audio steps separately: $0.01 per minute of audio to transcribe and about $0.036 to $0.038 per minute of speech to synthesize. Together that is about $0.046 to $0.048 for a minute of speech each way, before any LLM or phone line, and it is an offline job, not a live call.

Telnyx figures are from its Voice AI pricing page, read on 2026-10-03. Sume's per-minute TTS figure is my arithmetic: Sume TTS 1.0 is $0.0475 per 1,000 characters, and I assume 750 to 800 characters of script per spoken minute.

What the Telnyx page lists

The page lists one engine rate and then add-ons.

The arithmetic

Engine $0.05, plus Kimi at about $0.00644, plus inbound SIP at $0.0032, is $0.0596, which matches the page's roughly $0.06. At 10,000 minutes a month that is about $596.

If you bought the two audio steps from Sume instead, one minute of caller audio transcribed ($0.01) and one minute of agent speech synthesized (about $0.036 to $0.038) is $0.046 to $0.048. That is already close to Telnyx's $0.05 engine rate with nothing live, no LLM and no phone line. So for live calls the bundled rate is the better buy.

What Sume does and does not do

Sume has no phone number, SIP trunk or real-time audio stream, and no conversation loop. Both audio steps are jobs; see Jobs and results.

What it does well sits on either side of the call.

• Pre-render the lines that never change, so they are a stored file and not a per-call voice cost.
• Transcribe saved recordings after the call with word timings, for QA or captions.
• Make explainer or onboarding video that the call links to.

Volume table

Using the page's roughly $0.06 production figure and Sume's two audio lines added together (my arithmetic, $0.046 to $0.048 per minute of speech each way):

Reading the table

The Sume column is not a like-for-like alternative. It leaves out the LLM, the phone line and the live turn loop, and it has latency measured in a job, not in milliseconds. It shows only that buying the two audio steps separately costs about the same as the bundled engine rate, so any reason to prefer separate lines has to be about control or reuse, not price.

How to decide

Bundled pricing is easy to forecast: one engine rate plus an LLM and a line. Separate pricing is easier to optimize if only part of your audio is live. Count how many of your minutes are truly unscripted before deciding. The broader stack is covered in AI voice agent cost per minute.