On the occasion of the first anniversary of the Mistral 7B model, Mistral AI yesterday presented a new family of models, which it calls “Les Ministralaux”: Ministral 3B and Ministral 8B. These two models designed to run on peripheral devices, such as smartphones or laptops, offer unprecedented performance in the category of AI models with less than 10 billion parameters, meeting the growing needs for efficiency, calculation local and privacy protection.
Ministrals both have a pop-up window of up to 128,000 tokens, Ministral 8B benefits from an optimized attention architecture for faster, memory-efficient inference. According to Mistral AI, they set a new frontier in knowledge, common sense, reasoning, function calling and efficiency in the category.
In its press release, the unicorn declares:
“Our most innovative customers and partners are increasingly demanding local, privacy-aware inference for critical applications such as on-device translation, internet-free smart assistants, local analytics, and autonomous robotics. . Ministrals were designed to provide a computationally efficient, low-latency solution for these scenarios. From independent hobbyists to global manufacturing teams, Ministrals meet a wide variety of use cases.”
Used in combination with larger models like Mistral Large, Ministrals act as efficient intermediaries for executing complex tasks, such as function calls in multi-step workflows.
Performance comparisons
Ministrals were tested on various tasks where they performed better than their peers: Gemma 2 2B, Llama 3.2 3B, Llama 3.1 8B and Mistral 7B. Mistral AI clarifies that all models have been re-evaluated using its internal framework for fair comparison purposes.
Pre-trained models
The Ministral 3B and 8B Instruct models were also compared to Gemma 2 2B, Llama 3.2 3B, Llama 3.1 8B, Gemma 2 9B and Mistral 7B on different evaluation categories.
Availability and pricing
Both models are already available under the Mistral commercial license, Ministral 8B can be downloaded but only for research purposes.
Developers can use Ministral 3B and Ministral 8B via Mistral AI’s cloud platform, La Plateforme, and soon on other partner clouds.
Ministral 8B is priced at $0.1 per million output/input tokens, while Ministral 3B comes in at $0.04 per million output/input tokens.
Related News :