Skip to main content

​
Mistral Provider

The Mistral provider gives you access to Mistral AI’s models, including general-purpose language models, vision models, and specialized coding models.

​
Configuration

​
API Key Setup

Set your Mistral API key in your .env file: 
MISTRAL_API_KEY=your_api_key_here
The API key configuration is defined in config/llm-magic.php: 
'apis' => [
    'mistral' => [
        'token' => env('MISTRAL_API_KEY'),
    ],
]

​
Available Models

LLM Magic supports the following Mistral models:

​
Language Models

  • mistral-large-latest - Most powerful Mistral model
  • mistral-medium-latest - Balanced performance
  • mistral-small-latest - Fast and efficient
  • mistral-saba-latest - Specialized Saba model

​
Vision Models (Pixtral)

  • pixtral-large-latest - Large vision model
  • pixtral-12b-2409 - 12B parameter vision model

​
Code Models

  • codestral-latest - Specialized coding model

​
Compact Models (Ministral)

  • ministral-3b-latest - 3B parameter compact model
  • ministral-8b-latest - 8B parameter compact model

​
Research Models

  • open-mistral-nemo - Mistral Nemo research model
  • open-codestral-mamba - Codestral Mamba research model

​
Model Constants

Use these constants for type safety: 
use Mateffy\Magic\Models\Mistral;

Mistral::MISTRAL_LARGE          // mistral-large-latest
Mistral::MISTRAL_MEDIUM         // mistral-medium-latest
Mistral::MISTRAL_SMALL          // mistral-small-latest
Mistral::MISTRAL_SABA           // mistral-saba-latest
Mistral::PIXTRAL_LARGE          // pixtral-large-latest
Mistral::PIXTRAL_12B            // pixtral-12b-2409
Mistral::CODESTRAL              // codestral-latest
Mistral::MINISTRAL_3B           // ministral-3b-latest
Mistral::MINISTRAL_8B           // ministral-8b-latest
Mistral::MISTRAL_NEMO           // open-mistral-nemo
Mistral::CODESTRAL_MAMBA        // open-codestral-mamba

​
Usage

​
Using the Constructor

Create a Mistral model instance: 
use Mateffy\Magic\Models\Mistral;
use Mateffy\Magic\Models\Options\ChatGptOptions;

$model = new Mistral(
    model: Mistral::MISTRAL_LARGE,
    options: new ChatGptOptions
);

​
Getting Available Models

Retrieve a list of all available Mistral models: 
use Mateffy\Magic\Models\Mistral;

$models = Mistral::models();
// Returns a Collection with prefixed model names like 'mistral/mistral-large-latest'

$models = Mistral::models(prefix: null, prefixLabels: null);
// Returns models without prefix
The models collection includes human-readable labels: 
[
    'mistral/mistral-large-latest' => 'Mistral Large',
    'mistral/mistral-medium-latest' => 'Mistral Medium',
    'mistral/mistral-small-latest' => 'Mistral Small',
    'mistral/pixtral-large-latest' => 'Pixtral Large',
    'mistral/pixtral-12b-2409' => 'Pixtral 12B',
    'mistral/codestral-latest' => 'Codestral',
    'mistral/ministral-3b-latest' => 'Ministral 3B',
    'mistral/ministral-8b-latest' => 'Ministral 8B',
    'mistral/open-mistral-nemo' => 'Mistral Nemo (Research)',
    'mistral/open-codestral-mamba' => 'Codestral Mamba (Research)',
    'mistral/mistral-saba-latest' => 'Mistral Saba',
]

​
API Configuration

Mistral uses the OpenAI-compatible API with a custom base URI: 
protected function getOpenAiBaseUri(): ?string
{
    return 'https://api.mistral.ai/v1';
}

​
Organization Info

The Mistral provider includes organization metadata:
  • ID: mistral
  • Name: Mistral
  • Website: https://mistral.ai
  • Privacy: Data may be used for model training and abuse prevention

​
Vision Support

Pixtral models support vision capabilities with image inputs.

​
Image Limit

Mistral models are limited to 8 images per request: 
$model = new Mistral(Mistral::PIXTRAL_LARGE);
$maxImages = $model->getMaximumImageCount(); // Returns 8
This limit is enforced for all Mistral models implementing the HasMaximumImageCount interface.

​
Model Selection Guide

​
General Language Tasks

  • Mistral Large: Most powerful, best for complex reasoning
  • Mistral Medium: Balanced performance and cost
  • Mistral Small: Fast and cost-effective for simple tasks

​
Vision Tasks

  • Pixtral Large: Best vision capabilities
  • Pixtral 12B: Faster vision processing

​
Code Generation

  • Codestral: Optimized for code generation and completion
  • Codestral Mamba (Research): Experimental coding model

​
Edge/On-Device

  • Ministral 3B: Smallest model for resource-constrained environments
  • Ministral 8B: Better performance while still compact

​
Research

  • Mistral Nemo: Open research model for experimentation
  • Codestral Mamba: Research model with Mamba architecture

​
Advanced Options

You can pass additional options using ChatGptOptions: 
use Mateffy\Magic\Models\Mistral;
use Mateffy\Magic\Models\Options\ChatGptOptions;

$options = new ChatGptOptions(
    // Configure temperature, max tokens, etc.
);

$model = new Mistral(Mistral::MISTRAL_LARGE, $options);

​
Example Usage

​
Text Generation

use Mateffy\Magic\Models\Mistral;

$model = new Mistral(Mistral::MISTRAL_LARGE);
// Use the model for chat completions

​
Vision with Pixtral

use Mateffy\Magic\Models\Mistral;

$model = new Mistral(Mistral::PIXTRAL_LARGE);

// Remember: Maximum 8 images per request
if (count($images) <= $model->getMaximumImageCount()) {
    // Process images with the model
}

​
Code Generation

use Mateffy\Magic\Models\Mistral;

$model = new Mistral(Mistral::CODESTRAL);
// Use for code completion and generation tasks

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us