Ted Hisokawa
October 16, 2024 14:14
Mistral AI launches Ministral 3B and 8B, cutting-edge models for edge computing. These models promise low latency and efficient performance for a variety of applications.
According to Mistral AI, Mistral AI has announced the launch of two new models, Ministral 3B and Ministral 8B, designed specifically for on-device computing and edge use cases. This model was released on the one-year anniversary of Mistral 7B, marking an important milestone in cutting-edge AI innovation.
Advanced features and use cases
Ministral models are designed to excel in areas such as knowledge, common sense reasoning, function calling, and efficiency within the lower 10B category. Ministral 8B features a unique interleaved sliding window attention pattern for improved speed and memory efficiency and offers a wide range of context lengths up to 128k. These features make the model suitable for a wide range of applications, including on-device translation, internet-free smart assistants, local analytics, and autonomous robotics.
Working with large-scale language models like Mistral Large, Ministral models can act as efficient intermediaries in complex workflows, parsing input, routing operations, and calling APIs with minimal latency and cost. This makes it an ideal solution for both independent developers and large manufacturing teams seeking a privacy-first, low-latency inference solution.
Performance and Benchmarks
Mistral AI benchmarked the performance of the Ministral 3B and 8B against other models, including the Gemma 2 2B, Llama 3.2 3B, and Mistral 7B. The results show that the Ministral model consistently outperforms its peers on a variety of tasks. These evaluations highlight the model’s ability to efficiently handle diverse and complex scenarios.
Availability and Pricing
Both models are currently available, with prices set at $0.1 per million tokens for Ministral 8B and $0.04 per million tokens for Ministral 3B. This model is available under Mistral’s commercial and research licenses and includes self-deployment options under commercial licenses and support for lossless quantization to optimize performance for specific use cases. Additionally, the model weights in the Ministral 8B Instruct are accessible for research purposes.
future prospects
Mistral AI continues to innovate on frontier AI models as we strive to push the boundaries of what is possible in edge computing. Since the launch of the Mistral 7B, the company has made significant progress, as evidenced by the outstanding performance of the new Ministral 3B model. Mistral AI looks forward to receiving feedback from users as it explores the capabilities of the Ministral model.
Image source: Shutterstock