ModelSheep
All models

Voxtral Small 24B API

Send a recording and a question to Voxtral Small 24B from the OpenAI SDK. Read the transcript, answer, or summary from the reply. From $0.090 per million input tokens.

Get API keyRead the docs
  • Audio and text
  • 33K token context
  • 24.3B parameters
  • Apache 2.0 licence
from openai import OpenAI

client = OpenAI(
    base_url="https://api.modelsheep.com",
    api_key=API_KEY_MODELSHEEP,
)

response = client.chat.completions.create(
    model="mistralai/Voxtral-Small-24B-2507",
    messages=[
        {
            "role": "user",
            "content": "Why do sheep count people?"
        }
    ],
)

print(response.choices[0].message.content)

How to use the Voxtral Small 24B API

  1. 1

    Create a ModelSheep account and copy an API key from the dashboard.

  2. 2

    Point the OpenAI SDK at the base url https://api.modelsheep.com and set the model to mistralai/Voxtral-Small-24B-2507.

  3. 3

    Send one user message that holds a text part and an audio part, and read the transcript, answer or summary back from the first choice.

What Voxtral Small 24B is

Voxtral Small 24B is a speech understanding model from Mistral AI. It adds audio input to Mistral Small 3 while keeping its text ability. One call replaces two models: a speech recogniser and a language model. Mistral ships a 3B build for edge devices; this build handles production traffic.

What it does well

  • Transcribes a recording and detects the spoken language without a language flag.
  • Answers a question about a recording or summarises it without a separate transcription step.
  • Converts a spoken instruction into a function call for your backend.

What it is not for

  • Long recordings. Mistral caps transcription at 30 minutes of audio and audio understanding at 40 minutes. Split longer files.
  • Speech output. The model reads audio and writes text. It never returns audio.

Calling it

Post to /chat/completions and name mistralai/Voxtral-Small-24B-2507 as the model. The audio rides inside the message content as an audio part, next to your text, so a caller who sends a file upload or a bare URL gets nothing back. Mistral encodes that part with the mistral-common library and runs audio understanding at temperature 0.2.

mistralai/Voxtral-Small-24B-2507 /chat/completions

SDKs and languages

Voxtral Small 24B answers on an HTTP endpoint, so any language that can post JSON reaches it. These are the ones the box above writes out.

OpenAI SDK
PythonTypeScriptGoJavaC#PHPRust
Anthropic SDK
PythonTypeScriptGoJavaC#PHPRust
HTTP
cURLPythonTypeScriptGoRustJavaC#PHP

Specifications

AuthorMistral AI
Categoryaudio
Acceptstext, audio, file
Returnstext
Total parameters24.3B
Active parametersdense, all of them
Scorenot rated
Languages8 namedEnglish, Spanish, French, Portuguese, Hindi, German, Dutch and Italian. Mistral writes "to name a few", so eight is a floor. The claim covers speech, not the text side.
Precision servedFP8
Context33K tokens
Vectorn/a
LicenseApache 2.0

Price

Input

$0.090

per million tokens

Output

$0.270

per million tokens

Against OpenRouter

10%

below the cheapest of 3 providers

The cheapest OpenRouter price for this model was $0.100 in and $0.300 out, with a median of $0.100 and $0.300 across the comparable providers. Read on 2026-09-17.

License

Mistral publishes the weights under Apache 2.0. Commercial use needs no further permission, and no vendor terms sit on top of the licence.

Similar models

MiMo V2.5XiaomiMiMo/MiMo-V2.5
Parameters
310.8B
Context
1.1M
Input per 1M
$0.126
Output per 1M
$0.252
License
MIT
Faster Whisper Large v3Systran/faster-whisper-large-v3
Parameters
1.55B
Context
0K
Input per 1M
$0.024 /hr
Output per 1M
n/a
License
MIT
GLM 5.3 Flashzai-org/GLM-5.3-Flash
Parameters
321.3B, 18B
Context
1.3M
Input per 1M
$0.081
Output per 1M
$0.270
License
MIT

Start using Voxtral Small 24B

Point the OpenAI SDK at api.modelsheep.com. Your prompts stay in Germany, and nothing is retained.

Get API keyRead the quickstart