Mistral AI, the French AI startup, has released its first open-source large language model, Mistral 7B. According to the company, the new LLM performs better than other models of the same size and is unique in that it can be used for free without any restrictions.
Mistral 7B is a 7-billion-parameter model and the company envisions this model as an open-source rival to current AI solutions. The raw model weights are available on BitTorrent and Hugging Face, making it accessible for developers and businesses alike.
While also offering natural coding capabilities, the LLM is capable of generating text in over 100 languages, including English, Spanish, French, German, Chinese, Japanese, and many others. This makes it a versatile tool for multiple enterprise-centric use cases.
This model is released under the Apache 2.0 license and can be used by anyone from hobby users to multinational corporations. However, running it locally requires a corresponding system and may require paying for cloud resources.
What sets Mistral 7B apart from other “small” large language models like Llama 2 is its efficiency. It offers similar capabilities but at a significantly lower computational cost. Fundamental models like GPT-4 can do much more but are so expensive and complex to operate that they are only available through APIs or remote access. Mistral 7B significantly outperforms Llama 2 13B in all evaluation criteria and performs on par with Llama 34B.
Apart from that, they also offer Mistral 7B Instruct models. These models are designed to be highly adaptable, catering to specific tasks and user needs. You can even chat with the Mistral 7B Instruct model on Perplexity Labs, offering a hands-on experience for users.
Founded just six months ago, Mistral AI has already made waves with its record-breaking $118 million seed round, the largest in European history. The company was founded by alumni from Google DeepMind and Meta and aims to “make AI useful” for enterprises by leveraging publicly available data and customer contributions.
According to their pitch deck, Mistral AI aims to develop models that will surpass OpenAI’s offerings by 2024. With strong financial backing, the company plans to train models on larger infrastructures, positioning itself as a research leader in AI.