Mistral AI has announced the release of two new AI models, Ministral 3B and Ministral 8B, designed for edge computing. These models, collectively called "Les Ministraux," aim to provide efficient and low-latency solutions for on-device applications such as translation, smart assistants, local analytics, and robotics. The models are notable for their ability to handle a context length of up to 128,000 tokens, allowing them to process extensive information efficiently.
Key Features of Les Ministraux:
- Models: Ministral 3B and Ministral 8B
- Capabilities: High performance in knowledge, reasoning, and function-calling within the sub-10B model category
- Context Length: Supports up to 128k tokens
- Special Features: Ministral 8B includes an interleaved sliding-window attention pattern for improved inference speed and memory efficiency.
Availability and Pricing:
Both models are available immediately. Ministral 8B is priced at $0.1 per million tokens, while Ministral 3B costs $0.04 per million tokens. The model weights for Ministral 8B are available for research use, whereas a commercial license is required for Ministral 3B.
Company Background:
Mistral AI is a French AI startup known for its open-weight models that developers can customize and deploy. The company focuses on providing transparent and decentralized AI technologies. Mistral AI's offerings include serverless APIs and cloud services, ensuring flexibility and independence from major cloud providers.
Mistral AI
By introducing Les Ministraux, Mistral AI continues to advance its mission of delivering powerful AI solutions that prioritize privacy and efficiency, catering to both individual developers and large enterprises.
