pa5haw/Magistral-Small-2509-mlx-fp16

VISIONConcurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The pa5haw/Magistral-Small-2509-mlx-fp16 model is a 24 billion parameter language model, converted to MLX format from the Mistral AI Magistral-Small-2509 architecture. This model is specifically designed for efficient inference on Apple Silicon devices using the MLX framework, supporting a 32768 token context length. Its primary utility lies in local, high-performance language generation and processing tasks on compatible hardware.

Loading preview...

Model Overview

The pa5haw/Magistral-Small-2509-mlx-fp16 is a 24 billion parameter language model, derived from the mistralai/Magistral-Small-2509 architecture. This specific version has been converted to the MLX format using mlx-lm version 0.31.2, making it optimized for efficient inference on Apple Silicon devices.

Key Capabilities

  • MLX Optimization: Tailored for high-performance execution on Apple Silicon, leveraging the MLX framework.
  • Large Parameter Count: With 24 billion parameters, it offers robust language understanding and generation capabilities.
  • Extended Context Window: Supports a context length of 32768 tokens, enabling processing of longer inputs and generating more coherent, extended outputs.
  • Mistral AI Foundation: Built upon the Mistral AI architecture, known for its strong performance in various language tasks.

Good For

  • Local Inference: Ideal for developers and users who require powerful language models to run directly on their Apple Silicon hardware.
  • Application Development: Suitable for integrating advanced language capabilities into applications that benefit from local, optimized model execution.
  • Experimentation: Provides a robust base for experimenting with large language models in an MLX-native environment.