Compumacy/m3.1

VISIONPricing:Input $0.867 / Output $1.61Concurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Mar 20, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Compumacy/m3.1 is a 24 billion parameter base model developed by Mistral AI, building upon Mistral Small 3. It features state-of-the-art vision understanding and an enhanced long context capability of up to 128k tokens, while maintaining strong text performance. This model is designed for advanced text and vision tasks, offering multilingual support and an Apache 2.0 license for broad commercial and non-commercial use.

Loading preview...

Overview

Compumacy/m3.1 is a 24 billion parameter base model from Mistral AI, an evolution of Mistral Small 3. It significantly enhances capabilities by integrating state-of-the-art vision understanding and extending the context window to 128k tokens, all while preserving its strong text processing performance. This model serves as the foundation for the instruction-tuned Mistral-Small-3.1-24B-Instruct-2503.

Key Capabilities

  • Vision Understanding: Analyzes images and provides insights based on visual content, complementing text processing.
  • Extended Context Window: Features a 128k token context window, enabling processing of much longer inputs and generating more coherent, extended outputs.
  • Multilingual Support: Supports dozens of languages, including English, French, German, Japanese, Korean, Chinese, and many others.
  • Apache 2.0 License: Offers flexibility for both commercial and non-commercial applications.

Benchmark Highlights

In pretrain evaluations, Compumacy/m3.1 (Small 3.1 24B Base) demonstrates strong performance:

  • MMLU (5-shot): 81.01%
  • MMLU Pro (5-shot CoT): 56.03%
  • MMMU: 59.27%

Good For

  • Multimodal Applications: Ideal for tasks requiring both text and image analysis, such as visual question answering or image captioning.
  • Long-form Content Processing: Suitable for handling extensive documents, codebases, or conversations due to its large 128k context window.
  • Multilingual Deployments: Effective for applications targeting a global audience with its broad language support.
  • Foundation for Instruction Tuning: As a base model, it provides a robust starting point for further fine-tuning into specialized instruction-following models.

Finetunes
1 models