French startup Mistral has released its first family of edge-friendly generative AI models, “Les Ministraux,” designed for run on laptops and phones. The models can be used for basic text generation or paired with more capable models to complete tasks. Two models are available: Ministral 3B and Ministral 8B, both with a context window of 128,000 tokens.
The company’s innovative customers have requested local, privacy-first inference solutions for applications like on-device translation and autonomous robotics. Mistral’s models aim to provide efficient and low-latency solutions for these scenarios.
Ministral 8B is available for research purposes only, while Ministral 3B can be used through Mistral’s cloud platform or other partner clouds starting soon. The pricing of the two models varies, with Ministral 8B costing 10 cents per million output/input tokens and Ministral 3B costing 4 cents per million.
Mistral claims its models outperform comparable ones from Google, Meta, and itself on AI benchmarks, showcasing their potential for edge hardware. The company continues to expand its AI product portfolio with recent launches like a free service for developers, an SDK, and new models like Codestral.
Founded by alumni from Meta and Google’s DeepMind, Mistral aims to create flagship models rivaling the best-performing ones in the market, including OpenAI’s GPT-4o and Anthropic’s Claude. The startup reported generating revenue this summer and is making progress towards its mission of creating profitable generative AI products.
Source: https://techcrunch.com/2024/10/16/mistral-releases-new-ai-models-optimized-for-edge-devices/