Mistral-Small-3.2-24B-Instruct-2506

Visual LLM

Baseado no Mistral Small 3 (2501), o Mistral Small 3.2 (2506) acrescenta uma compreensão de visão de última geração e melhora as capacidades de contexto longo até 128k tokens sem comprometer o desempenho do texto. Com 24 mil milhões de parâmetros, este modelo atinge capacidades de alto nível tanto em tarefas de texto como de visão. Este modelo é uma pequena atualização do Mistral-Small-3.1-24B-Instruct-2503, que é uma versão ajustada por instruções do: Mistral-Small-3.1-24B-Base-2503.

Experimente outro modelo

Sobre o modelo Mistral-Small-3.2-24B-Instruct-2506

Publicado em huggingface

20/06/2025

Licença: Apache 2.0

Preço de entrada

0.09 € /Mtoken(entrada)

Preço de saída

0.28 € /Mtoken(saída)

Funcionalidades suportadas

Chamada de funçãoMultimodalStreaming

Formatos de saída

raw_textjson_objectjson_schema

Tamanhos de contexto

128k

Parâmetros

24B

Experimente o modelo brincando com ele.

Example of how to use Mistral-Small-3.2-24B-Instruct-2506 Vision Language Model API in python

Introduction

The following examples walk you through the use of a VLM (Vision Language Model), able to take images and texts prompts and generate text. To send a multimodal input to the model, you have to use a content list that will contain the text prompt and the base64 encoded image.

A VLM encodes the image as embeddings and uses tokens to represent and process the image along the usual text tokens, so an image will use some of the input context length of the model.

AI Endpoints VLMs

These Vision Language Model APIs are based on Open-Source models:

Qwen2.5-VL-72B-Instruct: A versatile model that can handle both image and text input, as well as text-only requests
llava-next-mistral-7b: On the other hand, this model is specifically designed to process image and text input together. It cannot process text-only requests. Therefore, it should not be used as a simple LLM.

Please ensure that you choose the appropriate model based on your specific use case.

With a simple HTTP client (requests)

First, install the requests library:

pip install requests

Next, export your access token to the OVH_AI_ENDPOINTS_ACCESS_TOKEN environment variable:

export OVH_AI_ENDPOINTS_ACCESS_TOKEN=<your-access-token>

If you do not have an access token key yet, follow the instructions in the AI Endpoints – Getting Started.

If you don't have any images available for testing, save the following image locally as sample.jpg:

And then run the following example Python code to ask the VLM to describe this image (or perform another task by adapting the content field of the messages dictionary):

import mimetypes
import os
import requests
import base64

image_filepath = "sample.jpg"
with open(image_filepath, "rb") as img_file:
    image_data = img_file.read()

# detect MIME type (default to jpeg if unknown)
mime_type, _ = mimetypes.guess_type(image_filepath)
if mime_type is None:
    mime_type = "image/jpeg"

encoded_image = f"data:{mime_type};base64,{base64.b64encode(image_data).decode('utf-8')}"

url = "https://oai.endpoints.kepler.ai.cloud.ovh.net/v1/chat/completions"
payload = {
    "max_tokens": 512,
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "Describe this image."
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": encoded_image
                    }
                }
            ]
        }
    ],
    "model": "Mistral-Small-3.2-24B-Instruct-2506",
    "temperature": 0.2,
}
 
headers = {
    "Content-Type": "application/json",
    "Authorization": f"Bearer {os.getenv('OVH_AI_ENDPOINTS_ACCESS_TOKEN')}",
}
 
response = requests.post(url, json=payload, headers=headers)
if response.status_code == 200:
    # Handle response
    response_data = response.json()
    # Parse JSON response
    choices = response_data["choices"]
    for choice in choices:
        text = choice["message"]["content"]
        # Process text and finish_reason
        print(text)
else:
    print("Error:", response.status_code, response.text)

With the Python OpenAI library

The Mistral-Small-3.2-24B-Instruct-2506 API is compatible with the OpenAI specification.

First install the openai library:

pip install openai

Next, export your access token to the OVH_AI_ENDPOINTS_ACCESS_TOKEN environment variable:

export OVH_AI_ENDPOINTS_ACCESS_TOKEN=<your-access-token>

If you do not have an access token key yet, follow the instructions in the AI Endpoints – Getting Started.

Finally, run the following Python code:

import mimetypes
import os
import base64

from openai import OpenAI

url = "https://oai.endpoints.kepler.ai.cloud.ovh.net/v1"

client = OpenAI(
    base_url=url,
    api_key=os.getenv("OVH_AI_ENDPOINTS_ACCESS_TOKEN")
)

def multimodal_chat_completion(new_message: str, image_filepath: str = None) -> str:
    new_user_message = {
        "role": "user",
        "content": [
            {
                "type": "text",
                "text": new_message
            }
        ]
    }
    
    if image_filepath is not None:
        with open(image_filepath, "rb") as img_file:
            image_data = img_file.read()

        # detect MIME type (default to jpeg if unknown)
        mime_type, _ = mimetypes.guess_type(image_filepath)
        if mime_type is None:
            mime_type = "image/jpeg"

        encoded_image = f"data:{mime_type};base64,{base64.b64encode(image_data).decode('utf-8')}"
        image_content = {
            "type": "image_url",
            "image_url": {
                "url": encoded_image
            }
        }
        new_user_message["content"].append(image_content)
    
    history_openai_format = [new_user_message]

    return client.chat.completions.create(
        model="Mistral-Small-3.2-24B-Instruct-2506",
        messages=history_openai_format,
        temperature=0.2,
        max_tokens=1024
    ).choices.pop().message.content


if __name__ == '__main__':
    print(multimodal_chat_completion("Describe this image.", "sample.jpg"))

Model rate limit

When using AI Endpoints, the following rate limits apply:

Anonymous: 2 requests per minute, per IP and per model.
Authenticated with an API access key: 400 requests per minute, per Public Cloud project and per model.

If you exceed this limit, a 429 error code will be returned.

If you require higher usage, please get in touch with us to discuss increasing your rate limits.

Going Further

Want to explore the full capabilities of the LLM API? Dive into our dedicated Structured Output and Function Calling guides.

For a broader overview of AI Endpoints, explore the full AI Endpoints Documentation.

Reach out to our support team or join the OVHcloud Discord #ai-endpoints channel to share your questions, feedback, and suggestions for improving the service, to the team and the community.

Mistral-Small-3.2-24B-Instruct-2506