Photo Credit: Alexandra_Koch from Pixabay
In a market where Open AI and Anthropic remain the giants, there is a relatively new entrant in the space, that is causing quite a stir. Paris-based Mistral AI has been building an alternative to the large language models, and it is being rewarded with a significant market valuation.
Mistral AI’s OfferingsMistral AI was founded in April 2023 by Google’s DeepMind and Meta alumni, Arthur Mensch, Guillaume Lample and Timothée Lacroix. The co-founders met while students at École Polytechnique. It originally positioned itself as an AI company with an open-source focus to offer an alternative solution to proprietary-based offerings such as those of ChatGPT.Mistral has released a trio of open-source neural networks. Released earlier this year, it correctly answered 77.75% of the questions in, compared to the 79.5% score achieved by the most advanced edition of Meta Platforms Inc.’s new Llama 3 LLM. The highly efficient Mistral 8x22B is built on smaller neural networks that are each optimized for a different set of tasks. When prompted, Mixtral 8x22B activates the neural networks best suited to generate an answer, while keeping the rest dormant. This helps lower hardware requirements without compromising on its quality. Mistral 8x22B was released under the Apache 2.0 license, an open-source license that has no restrictions on use or reproduction beyond attribution.Mistral also offers a collection of paid, cloud-based LLMs, led by Mistral Large that also includes a calling feature that allows it to perform tasks in other applications. Developers can also package Mistral Large’s output into the JSON file format, to simplify it to make AI responses available through a company’s custom software. Mistral model, Mistral Large is available to users through a paid API with usage-based pricing. It currently costs $8 per million of input tokens and $24 per million of output tokens to query Mistral Large, which makes it almost 20% cheaper than the GPT-4 Turbo model.Mistral has already started exploring beyond the language model segment. One of its offerings – Mistral Embed – is an AI that the 79.5% score achieved by the most advanced edition of Meta Platforms Inc.’s new Llama 3 LLM.
Mistral’s FinancialsMistral is privately held so far and does not disclose financials. In fact, it does not even share how many users its services have.It has raised $750 million in funding so far from investors including Lightspeed Venture Partners, Andreessen Horowitz, Nvidia, Samsung Venture Investment Corporation, Salesforce Ventures, Belfius, Bertelsmann Investment, BNP Paribas, Bpifrance, Cisco, Eurazeo, Headline, Hanwha Asset Management’s venture fund, IBM, Korelya Capital, Latitude, Millennium New Horizons, Sanabil Investments, ServiceNow, and SV Angel. Its last round of funding was held earlier this year when it raised €600 million (~$640 million) at an estimated open-source license that has no restrictions on use or reproduction beyond attribution. Mistral plans to use the newly acquired funds to continue its global expansion as it focuses on getting AI into the hands of everyone.More By This Author:AI Unicorns: Icertis Gearing Up For An IPOAnalysis Of Google’s Character.ai AcquisitionAI Unicorns: Cyera Uses LLMs For Data Security
