French AI startup Mistral has released its first generative AI models designed to run on edge devices, like laptops and phones.
The new family of models, which Mistral calls “Les Ministrals,” can be used or tuned for a variety of applications, from basic text generation to working in conjunction with more capable models to accomplish tasks.
Two Les Ministralaux models are available – Ministral 3B and Ministral 8B – both of which have a pop-up of 128,000 tokens, meaning they can ingest approximately the length of a 50-page book.
“Our most innovative customers and partners are increasingly demanding local, privacy-focused inference for mission-critical applications such as on-device translation, internet-free smart assistants, local analytics, and autonomous robotics,” Mistral writing in a blog post. “The Ministrals were designed to provide a computationally efficient, low-latency solution for these scenarios. »
Ministral 8B is available for download today, but strictly for research purposes. Mistral asks developers and companies interested in Ministral 8B or Ministral 3B self-deployment configurations to contact it for a commercial license.
Alternatively, developers can use Ministral 3B and Ministral 8B through Mistral’s cloud platform, La Plateforme, and other clouds the startup has partnered with in the coming weeks. Ministral 8B costs 10 cents per million output/input tokens (~750,000 words), while Ministral 3B costs 4 cents per million output/input tokens.
There has been a recent trend toward smaller models, which are cheaper and quicker to train, fine-tune, and run than their larger counterparts. Google continues to add models to its Gem small family of models, while Microsoft offers its Phi collection of templates. In the latest update of its Lama sequel, meta introduced several small models optimized for cutting-edge hardware.
Mistral claims that Ministral 3B and Ministral 8B outperform comparable Llama and Gemma models, as well as its own. Mistral7B — on several AI benchmarks designed to assess instruction-following and problem-solving capabilities.
Mistral, based in Paris, which recently raised $640 million in venture capital, continues to gradually expand its AI product portfolio. In recent months, the company has spear a free service allowing developers to test its models, a SDK to allow customers to refine these models, and new ones modelsincluding a generative model of code called Codestral.
Co-founded by alumni of Google’s Meta and DeepMind, Mistral’s stated mission is to create flagship models that rival today’s most successful models, like that of OpenAI. GPT-4o and Claude from Anthropic – and ideally make some money in the process. Even if “making money” proves to be a challenge (as is the case for most generative AI startups), Mistral would have started generating revenue this summer.
