# Pricing
Source: https://docs.mserve.ai/docs/pricing
Summary: Prices in USD. Prepaid credits, no seats, no expiry.
Availability: available
Last reviewed: 2026-08-25

## Voice

Text to speech is billed per character of input text. Speech to speech is billed per second of input audio.

| Model | Model id | Price | Unit |
| --- | --- | --- | --- |
| Kokoro 82M | `kokoro-82m` | $3.00 per 1M characters | Input characters |
| Chatterbox | `chatterbox` | $6.00 per 1M characters | Input characters |
| Chatterbox Voice Conversion | `chatterbox-vc` | $0.060 per minute | Input audio, billed per second |

## Speech to text

Billed per minute of audio. Transcription and translation share the rate.

| Model | Model id | Price | Unit |
| --- | --- | --- | --- |
| Whisper Large v3 Turbo | `whisper-large-v3` | $0.006 per minute | Audio minutes |

## Real-time

Billed per token. Input and output tokens are priced separately, as in the `usage` block of every response. Reasoning tokens count as output tokens. Realtime WebSocket sessions are billed as audio minutes in, tokens, and characters out, at the rates on this page.

| Model | Model id | Input per 1M tokens | Output per 1M tokens |
| --- | --- | --- | --- |
| Qwen3.8 27B | `qwen3.8-27b` | $0.30 | $2.50 |

## Embeddings

| Model | Model id | Price per 1M tokens |
| --- | --- | --- |
| Qwen3 Embedding 0.6B | `qwen3-embedding-0.6b` | $0.010 |

## Batch

Billed per token. The 24h tier is the OpenAI `completion_window`. Flex is requested through `metadata: {"tier": "flex"}` and can take up to 72 hours.

| Model | Model id | 24h tier per 1M tokens | Flex tier per 1M tokens |
| --- | --- | --- | --- |
| Qwen3.8 27B | `qwen3.8-27b-batch` | $0.22 | $0.15 |

## Billing

- [Prepaid](/docs/resources/billing): Buy credits from $5 to $1,000 with a card. Each request debits the balance.

- **No seats**: One balance per account. Any number of keys, each with its own optional spend limit.

- **No expiry**: Credits stay until used.

- [Empty balance](/docs/resources/errors): Requests return 402 with code insufficient_balance. Nothing is queued.

> **Price changes**
> Every price has an effective date. Usage is billed at the price in effect when the request was made, so past invoices never change.