Vapi pricing per minute: $0.05 hosting plus pass-through lines

Vapi's usage-only plan charges $0.05 per minute for hosting, and its page lists a transcriber at $0.0095 to $0.0099, a model at $0.0077 to $0.0452 and a voice at $0.0146 to $0.0238 on top. Adding those lines to hosting gives about $0.082 to $0.129 per minute (my arithmetic), so hosting is roughly 39 to 61 percent of the bill. Sume is not a live call platform; the useful comparison is between Vapi's pass-through lines and the two audio lines Sume sells as files: speech-to-text at $0.01 a minute and text-to-speech at about $0.036 to $0.038 a minute of speech.

Everything about Vapi below comes from its pricing page, read on 2026-10-03. The speaking-rate figure that turns Sume's per-character price into a per-minute price comes from Cartesia's pricing page, which says a minute of audio generation takes 750 to 800 credits; Sume TTS uses about one credit per character, so this post reads that as 750 to 800 characters a minute.

What the Vapi page lists

The page prices the platform and then shows the component vendors at cost, with no markup. These are the numbers it gave for the default calculator selections.

Where the fee sits

Adding the lowest figures to hosting gives $0.0818 and the highest gives $0.1289, before transport. At the low end, hosting is 61 percent of the minute. At the high end, the model dominates and hosting is 39 percent. The practical reading: a cheaper model barely moves the total when the hosting fee is the biggest line.

The page also lists a Pro package at 10 percent of the Vapi hosting fee with a $999 monthly minimum, including 30 concurrent calls and 10 phone numbers. Check the plan terms before modelling it.

What Sume's lines cost beside it

Sume sells two audio steps as jobs. Speech-to-text is $0.01 per minute of audio, up to 600 seconds per job. Text-to-speech is $0.0475 per 1,000 characters, up to 20,000 characters or 1,200 seconds of audio per request; at 750 to 800 characters a minute that is about $0.036 to $0.038 per minute of speech. Each quote rounds up to whole cents.

Set against the Vapi lines, the STT price is close to Vapi's Deepgram pass-through ($0.0095 to $0.0099). The TTS price is higher than the ElevenLabs line Vapi shows ($0.0146 to $0.0238), so Sume is not the cheaper way to speak live.

What Sume does and does not do

Sume does not answer phone calls, hold a websocket open, or run a conversation. Both audio steps are asynchronous jobs: you submit, poll or wait on a webhook, and fetch a file. See Jobs and results.

Where it fits next to a Vapi build is the part that does not need to be live.

• Render a fixed greeting, hold message or voicemail drop once as an audio file, instead of paying a voice line on every call. See pre-rendering greetings.
• Transcribe recorded calls after the fact, in parts of up to 10 minutes, with word timings. See splitting long recordings.
• Make the marketing video around the product, which the platform does not do.

Monthly totals from those lines

Multiplying the low and high sums by call volume shows how quickly the hosting line scales. These are my arithmetic on the page's per-minute figures and exclude transport and any plan minimum.

Where a pre-rendered file helps

Suppose each call opens with a 10-second greeting. At Vapi's voice line of $0.0146 to $0.0238 a minute, that is $0.0024 to $0.0040 per call, or about $24 to $40 for 10,000 calls. Rendering the same greeting once as a file with Sume costs one cent for a 200-character script, because quotes round up to whole cents. The file still plays inside a hosted call, so the hosting minute is still billed; the saving is only the voice line.

That is a small saving next to the $500 hosting line at the same volume, which is why the platform fee, not the voice, is the number to negotiate first.

How to decide

If most of your minutes are live conversation, price the platform fee first, because it is the largest single line in Vapi's own calculator. If a large share of your audio is scripted and repeats, pre-rendering it removes that share from the live meter. For the stack arithmetic in general, see AI voice agent cost per minute.