> ## Documentation Index
> Fetch the complete documentation index at: https://developers.telnyx.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Meeting Pricing
> Pricing for the Telnyx Meeting API, billed per minute of bot time in a meeting.
A meeting session bills two things and nothing else. Everything beyond that is something you switch on, and each one adds its own line -- including one that is not a Telnyx charge at all. This page is split accordingly: what a plain session costs, then what each addition costs on top.
## Included by Default
Send a bot to a meeting and two things accrue, on every session, for as long as the bot is in the room.
**Bot presence** is the bot's time in the meeting. It is billed for every minute, or partial minute, from the moment it starts joining -- including time spent waiting for admission -- until it leaves, whether it speaks, transcribes, or just sits there.
**Transcription** is the live speech-to-text of what is said, billed per minute of audio transcribed at the [Speech-to-Text](/docs/voice/stt/overview) rate for the model in use -- Deepgram Nova 3 by default -- and it runs for as long as the bot is in the meeting.
| Usage | Price |
| - | - |
| Bot presence | `$0.0126 / min` |
| Transcription | `$0.0074 / min` |
| **Total** | **`$0.02 / min`** |
The transcription figure is the **Deepgram Nova 3** rate -- the model a meeting session uses by default.
Carried at no additional charge:
* **Transcript & artifact storage.** Not billed — keeping the transcript and any artifacts costs nothing. Only *generating* an artifact costs anything -- see [Artifacts](#artifacts) below. (Live transcription itself is billed, at the rate above.)
* **Recording**, which happens automatically, and its storage for the first **7 days**. Keeping it longer is charged -- see [Recording retention](#recording-retention).
* **Session events, webhooks and the MCP server.** Reading what happened costs nothing.
* **Chat.** Sending or receiving messages in the meeting's native chat is not billed.
* **The [Calendar API](/docs/meeting/join-meeting/calendar-auto-join).** Connecting a calendar, managing connections, listing tracked meetings and deciding per-meeting whether a bot joins are all free -- eight routes, no charge. Only a session a bot actually joins bills, at the rates above.
* **Sending, scheduling and removing bots.** The session lifecycle itself is not a line item.
## Additional Integrations
Each of these is billed separately from the two rates above -- nothing here is folded into the per-minute price. Every one except recording retention is off unless you ask for it; **recording retention is the opposite** -- recording happens automatically, so its storage begins billing after the free 7 days unless you delete the recording. You opt out of it, not into it.
| Addition | Price |
| - | - |
| Speech (TTS) | `$0.000003 / character` |
| AI Assistant | `$0.05 / min` + LLM tokens |
| Digital avatar | BYOK -- bring your own key |
| Artifacts | Inference token rates, per request |
| Recording retention past 7 days *(automatic -- opt out by deleting)* | `$0.05 / recording media hour / 30 days` |
### Speech
Text the bot speaks is billed per character at the rate for the configured `voice`, not per minute. A session that never calls `speak` and has no `speak_on_enter` incurs no speech charge at all, so this line varies far more between sessions than presence or transcription do.
Speech is currently metered as **raw text-to-speech usage**, separately from the per-minute presence meter. The intended direction is to fold an allowance into the presence rate and charge only for overage beyond it, which would change how this line appears on an invoice. Treat the per-character rate as accurate for what is billed today rather than as the settled model.
The default is **Telnyx TTS** at `$0.000003 / character` -- roughly `$0.003` for a thousand characters, or about 150 spoken words. Any Telnyx-billed voice can be selected and rates vary between them, but **you cannot bring your own text-to-speech provider**: the bring-your-own-key engines listed on the Text-to-Speech pricing page are not available to a meeting session.
### AI Assistant
A session that attaches an `assistant` is billed the advertised voice-agent rate of `$0.05 / min` for the duration the assistant is attached. LLM tokens are a separate line item at Telnyx GPU token rates, and frontier models from Anthropic and OpenAI pass through at their own rates.
**An assistant session bills presence, plus the assistant rate in place of transcription and speech.** The `$0.05 / min` rate already includes speech-to-text and text-to-speech, so those two rows do not also apply while an assistant is attached -- only presence continues alongside it.
No Call Control or telephony charges apply. The assistant reaches the meeting through the Meeting service directly rather than by bridging a call, so the per-minute carrier rates on the AI Assistants pricing page are not incurred here.
Switching `audio_gate` to `full_duplex` adds a per-participant surcharge on top of the rate above -- see [Audio Gate](/docs/meeting/ai-assistants#assistant-parameters) for what it changes and why it costs more.
### Digital Avatars
A [digital avatar](/docs/meeting/digital-avatars) is currently BYOK -- bring your own key. It requires **your own Anam account**, and the `api_key` you supply on the session mints a session token against that account, so the avatar's cost is not a Telnyx charge at all -- it arrives on an Anam invoice Telnyx never sees and cannot show you here.
Two things follow that are easy to miss:
* **Telnyx speech still applies.** Telnyx TTS drives the avatar's speech and lip sync, so everything the avatar says is billed under [Speech](#speech) as well.
* **A session with an avatar and an assistant is billed on three fronts at once** -- presence, the assistant rate plus its tokens, and Anam -- against a single meeting.
### Artifacts
Artifacts are not part of any per-minute rate. Each request -- `summary`, `action_items`, `decisions`, `topics`, `open_questions`, or a `custom` question -- is a separate inference run on Telnyx inference infrastructure, billed at [Inference API](https://telnyx.com/pricing/inference-api) token rates for the tokens it consumes. **That is the entire charge** -- storing and re-fetching the result afterward is free; see [Included by Default](#included-by-default) above.
Three consequences worth planning around:
* **`summarize_on_end` is a billed request.** It is convenient, not free -- a session that ends with it enabled has asked for one artifact.
* **The create call is not idempotent.** Two POSTs for the same type produce two artifacts and two charges; the service will not collapse them. Guard the call in your own code.
* **Cost scales with transcript length, not meeting length.** Generation reads up to the first 10,000 transcript segments, so a long, dense meeting costs more per artifact than a short one -- and asking six questions of one meeting is six runs over that same transcript.
### Recording Retention
Recordings are retained **free for the first 7 days**. After that, storage is `$0.05` per recording media hour for every additional 30 days, prorated hourly at `$0.000069` per media hour stored.
Two things about that unit are easy to misread. It is charged on the **length of the recording**, not the size of the file -- a one-hour recording is one media hour whatever its resolution. And it is multiplied by **how long you keep it**, so the cost of a given recording keeps accruing until it is deleted.
| Recording | Kept | Billed period | Cost |
| - | - | - | - |
| 1 hour | 7 days | none | `$0.0000` |
| 1 hour | 8 days | 1 day | `$0.0017` |
| 1 hour | 14 days | 7 days | `$0.0117` |
| 1 hour | 37 days | 30 days | `$0.0500` |
| 2 hours | 67 days | 60 days | `$0.2000` |
Deleting a recording stops the meter -- `DELETE /v2/meeting_sessions/{id}/recording_media`. If you only need the transcript, download the recording inside the free window and delete it.
## Example
A transcribe-only meeting -- no assistant, and the bot never speaks -- costs `$0.02 / minute` at the default **Deepgram Nova 3** model (`$0.0074` transcription + `$0.0126` presence):
| Meeting length | Calculation | Bill |
| - | - | - |
| 60 minutes | `60 * $0.02` | `$1.20` |
| 45 minutes | `45 * $0.02` | `$0.90` |
| 30 minutes | `30 * $0.02` | `$0.60` |
| 15 minutes | `15 * $0.02` | `$0.30` |
That is the whole bill for a session with nothing switched on. Every line in [Additional Integrations](#additional-integrations) is added on top: speech per character when the bot talks, the assistant rate in place of transcription on assistant sessions, one inference charge per artifact including a `summarize_on_end` summary, and -- on a separate invoice you will not see here -- Anam, if the bot wears an avatar.