Summary
The French AI company Mistral AI presents Ministral 3B and 8B, two new language models that were specifically developed for use on end devices and in edge computing scenarios.
Sea Mistral The Ministral 3B and Ministral 8B models, known as “Ministraux”, are currently the most powerful AI systems in their class for use on end devices and in edge computing environments. According to the company, both models support context lengths of up to 128,000 tokens.
Diverse application possibilities from translation to robotics
The new models target use cases where local processing and data protection are a priority. According to Mistral AI, the Ministraux models are particularly suitable for tasks such as on-device translation, internet-free intelligent assistants, local data analysis and autonomous robotics.
In combination with larger language models such as Mistral Large, the Ministraux should also be able to serve as efficient mediators for function calls in multi-stage workflows.
Advertisement
THE DECODER Newsletter
The important AI news directly to your email inbox.
✓ 1x weekly
✓ free
✓ Can be canceled at any time
Ministral 8B shows the best results overall, especially in the areas of knowledge, common sense and multilingual skills. | Image: Mistral
Share article
Recommend our article
Split
According to the benchmarks presented by Mistral, Ministral 3B and 8B outperform comparable models such as Gemma 2 2B or Llama 3.1 8B in many categories. According to Mistral AI, the performance of the smaller Ministral 3B is particularly impressive, performing better in some benchmarks than its larger predecessor, the Mistral 7B, which is likely to be one of the most widely used open source models.
Mistral 8B clearly beats Mistral 7B in all benchmarks. However, Mistral doesn’t have it with the newer one Llama 3.2 11B compared, which is probably a little better, but a little bigger. Instead, Mistral used the older Llama 3.1 8B in the benchmark.
Availability and prices
The new models are available now. According to Mistral AI, Ministral 8B can be used via the API for $0.1 per million tokens, while Ministral 3B is available for $0.04 per million tokens. The company offers commercial licenses for use on its own systems.
For research purposes Mistral AI provides the model weights of Ministral 8B Instruct. The company says both models will soon be available through cloud partners such as Google Vertex and AWS.