Mistral AI Introduces Les Ministraux: State-of-the-Art Edge Models for On-Device Computing

On the one-year anniversary of the release of Mistral 7B, which significantly advanced independent frontier AI innovation, Mistral AI is proud to introduce two new state-of-the-art models designed specifically for on-device computing and edge use cases. These models, named les Ministraux-Ministral 3B and Ministral 8B-are poised to set a new standard in knowledge, commonsense reasoning, function-calling, and efficiency within the sub-10B category.

Technical Overview

Key Features:

• Context Length: Both models support up to 128k context length (currently 32k on vLLM).
• Attention Pattern: Ministral 8B features a special interleaved sliding-window attention pattern for faster and more memory-efficient inference.
• Use Cases: Ideal for local, privacy-first inference in critical applications such as on-device translation, internet-less smart assistants, local analytics, and autonomous robotics.

Use Cases

Les Ministraux are designed to address the growing demand for local, privacy-first inference. Here are some key use cases:

• On-Device Translation: Efficiently translate text without relying on cloud services.
• Internet-Less Smart Assistants: Provide robust assistance in environments with limited or no internet connectivity.
• Local Analytics: Perform real-time data analysis directly on the device.
• Autonomous Robotics: Enable robots to make decisions based on local data processing.

These models can also serve as efficient intermediaries for function-calling in multi-step agentic workflows. They can be fine-tuned to handle input parsing, task routing, and API calls based on user intent across multiple contexts at extremely low latency and cost.

Benchmarks

To demonstrate the performance of les Ministraux, Mistral AI conducted a series of evaluations using their internal framework for fair comparison. Here are some highlights:

Pretrained Models

• Table 1: Ministral 3B and 8B models compared to Gemma 2 2B, Llama 3.2 3B, Llama 3.1 8B, and Mistral 7B across multiple categories.
• Figure 1: Graphical comparison of Ministral 3B and 8B base models against the aforementioned models.

Instruct Models

• Table 2: Ministral 3B and 8B Instruct models compared to Gemma 2 2B, Llama 3.2 3B, Llama 3.1 8B, Gemma 2 9B, and Mistral 7B on different evaluation categories.
• Figure 2: A comparison of the 3B family of Instruct models-Gemma 2 2B, Llama 3.2 3B, and Ministral 3B-showcasing the improvements of Ministral 3B over the much larger Mistral 7B.
• Figure 3: A comparison of the 8B family of Instruct models-Gemma 2 9B, Llama 3.1 8B, Mistral 7B, and Ministral 8B.

Availability and Pricing

Both Ministral 3B and Ministral 8B are available starting today. Here’s the pricing on la Plateforme:

| Model | API | Pricing on la Plateforme | License | |----------------|------------------------|--------------------------|-----------------------------------| | Ministral 8B | ministral-8b-latest | $0.1 / M tokens (input and output) | Mistral Commercial License |

Conclusion

Les Ministraux-Ministral 3B and Ministral 8B-are significant advancements in edge models, offering a blend of efficiency, performance, and privacy. Whether you're an independent hobbyist or part of a global manufacturing team, these models are designed to meet your on-device computing needs.