ETtech Explainer: All you need to know about OpenAI’s rival Mistral AI

Tech giant Microsoft has partnered with Paris-based generative artificial intelligence (AI) startup Mistral AI which would allow the latter’s AI models to run on Microsoft’s Azure cloud computing platform.

The less-than-year-old startup is making waves and being considered Europe’s response to AI platforms such as OpenAI’s ChatGPT and Google’s Gemini (formerly known as Bard)

ETtech tries to answer all your queries on Mistral AI


What is Mistral AI’s deal with Microsoft all about?
Mistral's latest model, Mistral Large, will be first available to Azure customers as part of the deal. Azure customers will be able to use several AI models from Mistral, which will also see its technology hosted on Microsoft's cloud computing platform.

Microsoft will acquire a minority stake in the Mistral AI, as per the Reuters report. It is important to note that Microsoft already owns a stake in Mistral's rival OpenAI after committing over $10 billion.

Does Mistral have such deals with other cloud providers?
Yes. In December last year, Mistral inked a similar deal with Microsoft’s rival Google to gain access to the latter’s cloud computing platform.

“As part of the agreement, Mistral AI will use Google Cloud's AI-optimized infrastructure, including TPU Accelerators, to further test, build, and scale up its LLMs (large language models), all while benefiting from Google Cloud's security and privacy standards,” both the companies had said in a joint statement.

Mistral has also been working with Amazon and other cloud platforms to distribute its latest model Mistral Large.

Tell us more about Mistral AI’s LLMs?
In December, Mistral quietly dropped the Mixtral 8x7B model in a post on X which claims to surpass the capabilities of Open AI’s GPT 3.5 and Meta’s Llama 2 large language models (LLMs) on several performance benchmarks.

“It is the strongest open-weight model with a permissive license and the best model overall regarding cost/performance trade-offs. In particular, it matches or outperforms GPT3.5 on most standard benchmarks,” Mistral AI said in a blog post.

In September, the company unveiled its first large language model, dubbed ‘Mistral 7B’ and has since become one of the fastest European startups to acclaim unicorn status.

Is Mixtral 8x7B model better than GPT 3.5 and Llama 2?
Besides surpassing several benchmarks on performance parameters, the Mixtral 8x7B model claims to use a far more cost-efficient method that allows models to be pre-trained with far less computing.

Although the model has 46.7 billion total parameters, it only uses 12.9 billion parameters per token. “It, therefore, processes input and generates output at the same speed and for the same cost as a 12.9B model,” the company said.

This can significantly reduce deployment costs for companies who are already running huge bills on data storage requirements as they scale up operations.

How much funding has Mistral AI raised so far?
Mistral raised $415 million in December in its second funding round in seven months, led by investors such as Andreessen-Horowitz and LightSpeed Ventures. Further, Nvidia and Salesforce have agreed to contribute about $130 million in convertible debt to the funding.

While the company did not disclose its valuation, a Reuters report pegged it to be at €2 billion, putting the French startup on the list of top AI companies in the world.

Some of its backers include Salesforce, BNP Paribas, General Catalyst, La Famiglia, Eric Schmidt, New Wave, Motier Ventures, and Sofina.

Who runs Mistral AI?
The French startup is run by CEO Arthur Mensch, Guillaume Lample, and Timothee Lacroix, who have previously worked with Meta and Google's DeepMind.