Mistral AI recently launched Mistral Small 3.1, an enhanced version of its Small 3 model. This advanced model boasts 24 billion parameters and has the capability to process both text and images with a context window of 128,000 tokens. According to the company, it outperforms similar models like Google Gemma 3 and GPT-4o Mini, while also being designed to operate efficiently on readily available hardware, such as an RTX 4090 graphics card or a Mac with 32 GB of RAM.
Amazon co-founder MacKenzie Scott has donated over $19 billion to charity in just five years
Diamond batteries powered by nuclear waste promise 28,000 years of clean energy
An AI Model Crafted for Efficiency
Mistral Small 3.1 is a multimodal and multilingual AI model developed to follow instructions, assist in conversations, analyze images, and handle complex tasks. With a processing speed of 150 tokens per second, this model is well-suited for applications requiring quick responses. Its extensive context window also enhances its ability to manage lengthy documents and perform advanced reasoning.
A notable feature of the model is its optimization for deployment on consumer-grade hardware. After being quantified, it can run locally on a single RTX 4090 graphics card or a Mac equipped with 32 GB of RAM, which is not standard but still significant. This design aims to eliminate the need for expensive cloud infrastructures.
An Open-Source Model under Apache 2.0 License
Continuing its commitment to open-source, Mistral AI offers Mistral Small 3.1 under the Apache 2.0 license. This approach allows businesses and developers to use, modify, and adapt the model with minimal restrictions. It contrasts with the strategies of some competitors who favor closed models available only through proprietary cloud services.
The model can be accessed through various platforms; it is available for download on Hugging Face in both base and instruct versions, or via Mistral AI’s API. It is also hosted on Google Cloud Vertex AI, with upcoming integrations planned for NVIDIA NIM and Microsoft Azure AI Foundry.
Strategically Positioning in European AI
Founded in 2023 by former researchers from Google DeepMind and Meta, Mistral AI is quickly establishing itself as a major player in the European AI landscape. Valued at $6 billion and having raised over a billion dollars, the company aims to position itself as a viable alternative to the American and Chinese giants in the industry.
Mistral AI’s dedication to open-source and model optimization for accessible hardware is part of a strategy aimed at offering AI solutions less reliant on costly infrastructures. This approach could help them attract a broader audience, from independent developers to large enterprises looking for alternatives to market-leading offerings.
It remains to be seen whether this open-source strategy will be sustainable in the long term against competitors with considerably greater financial resources. Nonetheless, Mistral Small 3.1 underscores Mistral AI’s commitment to delivering powerful and accessible AI models.