Mistral AI recently introduced two cutting-edge models, Ministral 3B and Ministral 8B, to commemorate the one-year anniversary of the launch of Mistral 7B. These models are specifically designed for on-device computing and edge applications, focusing on enhancing performance in critical areas like knowledge reasoning and function-calling.
Mistral AI has unveiled its latest models, Ministral 3B and Ministral 8B, in celebration of the success of Mistral 7B. The new models are tailored for on-device computing and edge applications to boost performance in key areas like knowledge reasoning and function-calling.
The Ministral models have the capability to handle up to 128k context length and utilize a unique sliding-window attention approach to ensure efficient inference in resource-constrained environments. These models cater to the rising demand for applications such as on-device translation, smart assistants, local analytics, and robotics, providing privacy-centric localized AI inference. Additionally, they act as intermediaries for larger AI models, enhancing task routing and API calling in various applications.
Mistral AI has responded to the increasing customer and partner demand for privacy-centric local inference by introducing the Ministral models. These models are specifically designed for critical applications like on-device translation, internet-less smart assistants, local analytics, and autonomous robotics, offering a compute-efficient and low-latency solution for diverse use cases.
Competing with counterparts like Gemma 2 2B, Llama 3.2 3B, and Llama 3.1 8B, the Ministral models have demonstrated impressive benchmarks. Both models, Ministral 3B priced at $0.04 per million tokens and Ministral 8B at $0.1, are now available for commercial use. Additionally, the weights for the 8B Instruct model are accessible for research purposes.
Mistral AI recently introduced Pixtral 12B, a model that combines text and image processing with 12 billion parameters. Utilizing vision encoding, Pixtral analyzes both images and text inputs, offering a robust multimodal AI solution.
In a move to compete with Meta’s Llama 3.1, Mistral AI launched Mistral Large 2, the latest version of its flagship model. This updated model enhances code generation, mathematics, multilingual support, and introduces advanced function-calling capabilities, now available on Mistral’s platform.
Mistral AI, based in Paris, has been expanding its portfolio following a successful $640 million venture capital raise. The company has introduced various services, including a free model-testing platform, an SDK for model fine-tuning, and a generative AI model for code named Codestral.
Founded by alumni from Meta and DeepMind, Mistral AI aims to develop flagship models that can rival leading AI systems like OpenAI’s GPT-4 and Anthropic’s Claude. While revenue generation remains a challenge for many AI startups, Mistral reported its first revenue growth earlier this year.
Sustainability and energy efficiency in AI
Amid concerns over the environmental impact of AI, companies like Mistral AI are focusing on sustainable development and energy-efficient models. The Ministral 3B and 8B models prioritize energy efficiency, emphasizing low-latency and compute-efficient inference to enable complex AI tasks on-device, reducing the need for energy-intensive cloud operations and benefiting edge applications.
(Photo by Amith Nair)
See also: Edge computing market set to surge to $378 billion by 2028
For more insights on edge computing, explore Edge Computing Expo happening in Amsterdam, California, and London.
Discover upcoming enterprise technology events and webinars powered by TechForge here.
