GPT-4 Getting New Competition from Mistral Large & Le Chat

French Startup Mistral AI comes close to beating ChatGPT with their new ‘Le Chat’. This is huge news in the world Generative AI world, so something is cooking here for sure. How good are these two latest tools? Let’s give you detailed insights about them without any further delay!

Highlights:

• Mistral AI releases Mistral Large and Le Chat, its latest large language model and conversational agent.
• Mistral Large is highly capable and comes close to GPT 4 in several benchmarks.
• Le Chat is a multilingual conversational assistant that lets you explore several models.

Looking into Mistral Large

Mistral AI has officially released Mistral Large, their latest flagship text generation model. Mistral Large is a highly capable LLM that can perform top-tier reasoning capabilities. It can perform highly complex multilingual reasoning tasks, including text understanding, transformation, and code production.

It has again made the headlines since its previous release of Mistral 8x7B in December when it beat several competitors including Meta’s Llama 2 34B and achieved a marvelous $2b valuation in 6 months.

It comes with several other features such as:

• Native fluency in English, French, Spanish, German, and Italian. It has an in-depth grammar sense and comprehensive context understanding.
• An impressive contextual token window having a token size of up to 32K. This allows for more input processing capabilities and an enhanced generation system.
• Developers will be able to personalize their own needs and moderate their usage policies accordingly as the model comes with precise instructions following capability.

But the most impressive feature has to be in the form of the function calling ability. This makes it possible to develop applications and modernize the tech stack at scale, in conjunction with the limited output mode that is used. Function calling enables more intricate interfaces with internal code, databases, or APIs by allowing developers to interface with Mistral endpoints with a suite of their tools.

February is filled with many such big AI announcements, including Google’s Gemma and Sora Text-to-Video AI.

Benchmarks: Does it stand toe-to-toe with GPT-4?

Mistral Large consists of impressive benchmarks mainly on commonly used factors. When looking at the aspect of MMLU (Massive Multitask Language Understanding), it ranks second to GPT-4 being the world’s second-ranked model available through an API.

Take a look at this model benchmark comparison published on their official website, where it ranks near GPT-4:

However, it still falls behind GPT-4 slightly when compared based on knowledge and reasoning. It surpasses Llama 2 70B and GPT 3.5 based on common sense reasoning and knowledge. The same trend is observed when compared to GPT-4 based on maths and coding skills. 

The question arising now is should developers make the switch from GPT-4 to Mistral Large? The benchmarks give the impression that it still has a bit of catching up to do.

However, the differences between the two are minimal and we shouldn’t overlook the fact that it is a new model in the market. It can be easily stated that Mistral quite holds the competition to GPT-4 compared to any other models before and if given time it won’t fail to impress.

So, for now, developers must enjoy its vast features surrounding function loading and multilinguistic understandings.

How can you access Mistral Large?

Mistral has partnered with Microsoft Azure intending to make Mistral Large ubiquitous with widespread access. This is a multi-year partnership that will allow Mistral AI to develop and deploy its models through Microsoft.

Here is what the company said about their official partnership:

“We are thrilled to embark on this partnership with Microsoft. With Azure’s cutting-edge AI infrastructure, we are reaching a new milestone in our expansion propelling our innovative research and practical applications to new customers everywhere.”

Mistral AI CEO Arthur Mensch.

You can access Mistral Large through Azure AI Studio and Azure Machine Learning, where you will get a seamless API-enabled experience.

It will also be accessible in Europe via Le Platforme which is safely hosted via Mistral’s infrastructure. With this access point, developers can create services and apps for their wide selection of models. 

CEO of Perplexity AI, Aravind Srinivas recently posted on X, that Mistral Large is available on Perplexity Pro’s platform. There it can also be set as a default model:

Mistral has overall enabled the deployment of Mistral Large on self-deployed models for development purposes with the most sensitive use cases with access to a vast variety of model weights.

Try Le Chat: the Multilingual Conversational Assistant

Mistral also released their conversational assistant chatbot, named Le Chat. Le Chat serves as a conversational interface for interacting with Mistral AI’s range of models. It provides an enjoyable and educational method to learn about Mistral AI’s technology.

In addition to providing customers with responses from Mistral Large, the conversational chatbot gives them access to their newest smaller models, Mistral Small and Mistral Next.

The chatbot has been released in the beta phase but has still been developed to adhere to its policies of safety and ethics.

As developers you are free to try out the chatbot, however, expect less freedom and open availability of content explicitly when generating responses.

Conclusion

Both Mistral Large and Le Chat are excellent innovations in the world of Generative AI. With their outstanding features, they can be of much importance to the developer community. The big question now is how will both tools perform in the days to come. We are yet to see diverse user reviews across varied subjects and use cases on the tools.