> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.talkif.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.talkif.ai/_mcp/server.

# Credits and pricing

> How Talkif charges — what a call costs and why, the four kinds of balance and the order they're spent in, what happens at zero, and a worked example.

Talkif is prepaid and usage-based. You keep a USD balance; every call, phone number and post-call analysis takes what it costs from that balance, itemised. There are no plans to pick and no per-seat fees — the price of a call is the sum of the resources it used, at the rates for the providers *you* chose in the flow.

This page explains the model. Topping up, cards and auto-recharge are in [Credits and payment](/account/credits-and-payment); reading a bill is in [Cost breakdown](/account/cost-breakdown).

## What a call costs

A finished call is priced as a set of line items, one per resource, each with its own unit:

| Line item          | Unit                                                                                                                                                                       | How it's measured                                                                                           |
| ------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- |
| **Telephony**      | per minute, **rounded up** to the next whole minute                                                                                                                        | Carrier time for the call. Rate depends on provider (Talkif number, your SIP trunk, WhatsApp) and direction |
| **STT**            | per minute, prorated to the second                                                                                                                                         | Duration of the call, at the chosen speech-to-text model's rate                                             |
| **LLM**            | per token, input and output priced separately                                                                                                                              | Every turn re-sends the prompts and transcript, so tokens grow with call length and prompt size             |
| **TTS**            | per character synthesized                                                                                                                                                  | Only what the agent actually said                                                                           |
| **Infrastructure** | per minute: prorated to the second for calls that start until November 14, 2026; **rounded up** to the next whole minute for calls that start from November 15, 2026 (UTC) | The agent runtime itself — the platform fee, a flat platform rate                                           |
| **Function calls** | —                                                                                                                                                                          | Calls to your own HTTP endpoints are **free**; you pay only the LLM tokens around them                      |
| **LLM analysis**   | per token                                                                                                                                                                  | Post-call insights, if enabled — a separate line so you can see what analysis costs versus the conversation |

Two consequences worth knowing before you design a flow:

* **Telephony rounds up, and from November 15, 2026 so does the platform fee.** A 61-second call pays 2 telephony minutes and 61 seconds of STT. Its infrastructure line (the platform fee) is 61 seconds for calls that start until November 14, 2026, and 2 minutes for calls that start from November 15, 2026 (UTC). At the standard platform rate of \$0.05 per minute that is \$0.0508 before and \$0.10 after; a call that ends on a whole minute costs the same either way, and no call pays more than one extra minute of platform fee. Short calls are proportionally more expensive on the rounded lines — [WhatsApp calls](/telephony/whatsapp-business-calling) have no telephony line at all.
* **LLM cost is quadratic-ish in call length.** Each turn resends the whole context, so a long System Prompt is paid on every turn of every call. Keeping prompts short is a cost lever as well as a quality one — see [Write prompts](/build/write-prompts).

### Transferred calls

When an agent [transfers a call to a person](/build/transfer-calls-to-people), the AI part of the bill stops where the agent leaves the call:

* **STT, LLM, TTS and infrastructure** are billed for the time the agent was on the call, including a warm-transfer briefing — not for the conversation between the caller and your person afterwards.
* **Phone legs** are billed as **telephony**, at the rates below. The caller's leg is billed for the whole call, before and after the transfer. A transfer to a phone-number destination is an extra outgoing leg, billed from the moment it's answered at the rate of the **Call from** number's provider — ringing isn't billed.
* Time your person spends on the call in the **dashboard** carries no platform fee of its own.

Rates are published live. The current telephony and number prices are below; model rates are on [Models and voices](/reference/models-and-voices).

### Telephony

| Provider     | Inbound / min | Outbound / min |
| ------------ | ------------- | -------------- |
| SIP Trunking | \$0.0040      | \$0.0040       |
| Twilio       | \$0.0085      | \$0.0140       |
| WhatsApp     | Free          | Free           |

### Phone numbers

| Fee          | Amount          |
| ------------ | --------------- |
| **Purchase** | \$1.15 one-time |

Monthly cost varies by country and number type. Price shown is for US local numbers.

### Recording storage

| Fee         | Amount                |
| ----------- | --------------------- |
| **Storage** | \$0.5000 / GB / month |

Billed based on actual storage duration. Minimum 1 day charged.

## The four kinds of balance

Your balance is made of *buckets*, and which bucket pays for what is the part that surprises people.

**`subscription`** `expires`

Granted monthly by a subscription plan; expires at the end of the period. Spent **first**.

---

**`promotional`** `may expire`

Credits from a promotion, usually with an expiry date. Spent **second**.

---

**`bonus`** `never expires`

Welcome credit, referral rewards and similar. Spent **third**.

---

**`purchased`** `never expires`

Money you paid. Spent **last**, so it's never wasted while expiring credit exists.

---

Two rules sit on top of the order:

1. **Calls and analysis can be paid from any bucket.** Expiring buckets go first so you lose as little as possible.
2. **Phone number purchases and monthly rental can only be paid from purchased or subscription balance.** Bonus and promotional credit never buy or rent a number. This is why a new account with only its welcome credit can build and test but must top up before buying a line.

`GET /billing/balance` reports `available` (everything usable), `purchasedBalance`, `bonusBalance` and the `earliestExpiration` across buckets.

## Admission, overdraft and suspension

Talkif checks your balance **once per call, before it starts** — and never again during it.

* To start any call (inbound, outbound, browser test) the account must be active and have at least **\$1.00** available. Below that: outbound requests fail with `insufficient_balance`, inbound calls are answered busy, campaigns and schedules hold.
* Once a call is running it is **never cut off for money**. It's metered to the end and settled as one charge, even if that takes the balance negative.
* If the balance ends at or below zero, the account is **suspended**: no new calls until a top-up brings it positive. The dashboard, billing and your data stay fully accessible — suspension only blocks spending. A top-up reactivates the account automatically.
* With **auto-recharge** on, the top-up happens instead of the suspension: the shortfall triggers a charge to your default card for your configured amount.

## A worked example

You sign up, get the **\$5.00 welcome credit** (bonus), and add **\$50.00** by card (purchased).

```text
Purchased   $50.00
Bonus        $5.00
Available   $55.00
```

#### You buy a US local number at \$1.15/month

Number purchases can't use bonus, so the rental comes off *purchased* even though bonus is available.

```text
Purchased   $50.00  −  $1.15   =  $48.85
Bonus        $5.00             =   $5.00
Available                      =  $53.85
```

#### You make a 4 min 20 s test call

Telephony is billed per started minute; STT per second; TTS per character; the LLM per token. Infrastructure is per second for calls that start until November 14, 2026, and per started minute from November 15, 2026. Say the call comes to \$0.42:

```text
Telephony        5 min (started)  ×  $0.014/min   =  $0.07
STT              4.33 min         ×  model rate
LLM              ~2,800 tokens    ×  model rate
TTS              ~1,900 chars     ×  model rate
Infrastructure   4.33 min         ×  platform rate   (until Nov 14, 2026)
                 5 min (started)  ×  platform rate   (from Nov 15, 2026: ~$0.03 more at $0.05/min)
Call total                                        ≈  $0.42
```

Calls spend **expiring-first**, so the whole amount comes off bonus:

```text
Purchased   $48.85             =  $48.85
Bonus        $5.00  −  $0.42   =   $4.58
Available                      =  $53.43
```

#### Twelve more calls use up the bonus

Bonus hits \$0.00; the next call spends from purchased. Nothing else changes — you don't notice the bucket switch except in the transaction history, where each entry names its bucket.

#### A long campaign runs the balance to –\$0.80

The last call was admitted with \$1.10 available, ran long, and settled at \$1.90. A running call is always billed to the end, so the balance goes negative and the account is **suspended**: existing calls finish normally, the queue holds.

```text
Available    $1.10  −  $1.90   =  −$0.80   → suspended
Top-up      −$0.80  + $20.00   =  $19.20   → active, queue resumes
```

#### Next month, rental renews

\$1.15 is taken from purchased balance for the number. If the balance can't cover rental, the number's billing status moves to *past due* and eventually *suspended* — see [Buy a number](/telephony/buy-a-number).

## Where the numbers come from

Every line item on a bill records the **rate that applied at the time**, so historical bills don't change when prices do. Rates are per account — if you're on negotiated pricing, your rates apply automatically everywhere the public ones are shown here.

## Next

#### [Credits and payment](/account/credits-and-payment)

Top up, manage cards, set up auto-recharge.

#### [Cost breakdown](/account/cost-breakdown)

Read a call's bill line by line and find what drives cost.

#### [Models and voices](/reference/models-and-voices)

Current per-model rates for STT, LLM and TTS.

#### [Choose models and voices](/build/choose-models-and-voices)

Trade off latency, quality and cost per provider.