As its latest announcement shows, Paris-based AI startup Mistral AI is gradually building an alternative solution to OpenAI and Anthropic. The company is launching a new flagship large language model called Mistral Large. In terms of inference capabilities, it is designed to be comparable to other top models such as GPT-4 and Claude 2.
In addition to Mistral Large, the startup is also launching its own new service Le Chat, which replaces ChatGPT. This chat assistant is currently available in beta.
For those not familiar with Mistral AI, the company is best known for its capitalization statement, as it raised an exorbitant amount of funding in a very short period of time to develop its basic AI models. The company was officially incorporated in May 2023. Just a few weeks later, Mistral AI raised a $112 million seed round. In December, the company closed a $415 million funding round with Andreessen Horowitz (a16z) leading the round.
Founded by Google's DeepMind and Meta alumni, Mistral AI initially positioned itself as an open source-focused AI company. Mistral AI's first model was released under an open source license that gave access to model weights, but this is not the case for larger models.
Mistral AI's business model is increasingly similar to that of OpenAI, as the company offers Mistral Large through a paid API with usage-based pricing. Currently, querying Mistral Large costs $8 per million input tokens and $24 per million output tokens. In artificial language jargon, a token represents a small chunk of a word. For example, the word “TechCrunch” is split into two tokens: “Tech” and “Crunch” when processed by an AI model.
By default, Mistral AI supports a context window of 32,000 tokens (typically 20,000 words or more in English). Mistral Large supports English, French, Spanish, German, and Italian.
For comparison, GPT-4 with a 32,000-token context window currently costs $60 per million input tokens and $120 per million output tokens. Therefore, Mistral Large is currently 5-7.5 times cheaper than GPT-4-32k. The situation is changing rapidly, and AI companies update their prices regularly.
But how will Mistral Raj fare against GPT-4 and Claude 2? As always, that's very difficult to tell. Mistral AI claims that based on several benchmarks it ranks second after GPT-4. However, there may be differences in benchmark selection and actual usage. You'll need to look further into it to see how it performs in your tests.
Image credits: Mistral AI
Alternatives to ChatGPT
Mistral AI is also launching a chat assistant today called Le Chat. Anyone can sign up and try chat.mistral.ai. The company says it's currently a beta release and may have “bugs.”
Access to this service is free (for now) and users can choose from three different models. Mistral Small, Mistral Large, and a short and concisely designed prototype model called Mistral Next. Please also note that when using Le Chat, you do not have access to the web.
The company also plans to launch a paid version of Le Chat for enterprise customers. In addition to centralized billing, enterprise clients will be able to define moderation mechanisms.
Partnership with Microsoft
Finally, Mistral AI is also using today's news to announce a partnership with Microsoft. In addition to Mistral's own API platform, Microsoft plans to offer the Mistral model to Azure customers.
This is another model in Azure's model catalog, but it doesn't seem like a big deal. But it also means that Mistral AI and Microsoft are currently discussing opportunities for collaboration and more. The first advantage of this partnership is that Mistral AI is likely to attract more customers with this new distribution channel.
As for Microsoft, the company is a major investor in OpenAI's restricted interest subsidiary. But the company also welcomes other AI models on its cloud computing platform. For example, Microsoft and Meta have partnered to deliver Llama large-scale language models on Azure.
This open partnership strategy is a great way to keep Azure customers in the company's product ecosystem. It may also be useful with respect to anti-competitive surveillance.