The New Mistral AI Model Challenges OpenAI and Google

A new successful player has emerged in the world of artificial intelligence – the French startup Mistral AI, which recently introduced its flagship language model, Mistral Large. The company positions its development as a potential competitor to well-known AI products such as GPT-4 from OpenAI and Claude 2 from Anthropic. Additionally, the startup announced the launch of Le Chat, which aims to be an alternative to ChatGPT. It is currently available in beta format.

Mistral Large was designed to handle complex text processing tasks in multiple languages, including text comprehension, transformation, and code generation. One notable feature of this model is its ability to deeply understand the grammar and cultural context of different languages. This results in higher accuracy compared to counterparts from Google and OpenAI.

The Mistral Large context window is 32,000 tokens (more than 20,000 words in English). Thanks to this, it can effectively work with large volumes of text, reproduce information accurately, and follow instructions clearly. This opens up new opportunities for developers to customize moderation policies and incorporate specific features.

Despite the lack of 100% clarity regarding the actual performance of the model compared to other AI "heavyweights", Mistral AI is optimistic about the prospects of its development. According to MMLU tests, Mistral Large demonstrates an impressive accuracy of 81.2%, which is competitive with other models. For comparison, GPT-4 showed 86.4%, Gemini Pro 1.0 – 71.8%, and Llama 2 70B – 69.9%.

Mistral AI has stated that its product will be available through APIs and Azure AI, as the startup has entered into a partnership agreement with Microsoft. Additionally, its representatives have announced the release of a smaller version of the model, Mistral Small, and a chat application. They will help business teams visualize and evaluate the potential of new AI tools.