Mistral Collaborates with Microsoft to Launch a New Large AI Model Featuring Chat Application Integration

Mistral AI Launches Mistral Large and Partners with Microsoft

Mistral, the innovative AI startup known for its striking Word Art logo and record-breaking seed funding in Europe, has unveiled Mistral Large—its most extensive enterprise model to date. This launch is accompanied by a strategic partnership with Microsoft, providing Mistral with $16 million in new capital and enhanced distribution through Azure.

Mistral Large: Key Features

Available immediately, Mistral Large is a powerful text generation model adept at complex multilingual tasks, including text understanding, transformation, and code generation. According to the multitask language understanding (MMLU) benchmark, it ranks as the second-best model accessible via API, closely following GPT-4.

Primarily available through API and Azure AI, Mistral Large supports several languages, including English, French, Spanish, German, and Italian. While competitors like Google and OpenAI also offer multilingual models, Mistral claims its model provides a superior understanding of grammar and cultural nuances, resulting in enhanced performance.

With a context window of 32K tokens, Mistral Large excels in processing extensive documents and accurately recalling information. The model also offers precise instruction-following capabilities, allowing developers to customize their moderation policies and native function calls.

While comparisons to larger models like Gemini 1.5—capable of handling up to 1 million tokens—are ongoing, Mistral reports strong performance against rival models. In MMLU tests, Mistral Large achieved an accuracy of 81.2%, just behind GPT-4’s 86.4%. Notably, its language-specific performance outshone Meta's offerings.

However, Mistral Large has shown weaknesses in coding tasks; it recorded a 45.1% accuracy on the HumanE benchmark, falling behind GPT-3.5, GPT-4, and Gemini Pro 1.0.

Mistral Small Optimization and Distribution

In addition to Mistral Large, the company has released an optimized version of Mistral Small, aimed at enhancing latency and reducing costs. This model serves as an intermediary between Mistral's open-weight offerings and Mistral Large.

The partnership with Microsoft is pivotal for expanding Mistral's market reach. As part of this collaboration, Mistral's models will be accessible on Azure AI Studio and Azure Machine Learning. This positions Mistral as the second company to provide commercial language models on Azure. Azure users can utilize existing credits for seamless interaction with Mistral’s APIs, complemented by direct support access.

Arthur Mensch, co-founder and CEO of Mistral AI, stated, "At Mistral AI, we make generative AI ubiquitous—through our open-source models and by making our commercial models available where developers create. We are proud to announce the availability of Mistral Large on Azure AI. Microsoft’s trust in our model signifies progress in our journey to democratize frontier AI."

Future Prospects and Chat App Launch

Mistral is also set to partner with Amazon Web Services (AWS), allowing its open models on Amazon Bedrock, although a timeline for this integration has yet to be disclosed.

To foster trust and showcase potential applications, Mistral is introducing a chat app—a multilingual conversational assistant that enables teams to explore the capabilities of their models. Users can sign up for beta access at Mistral’s website, where they can engage with the AI in an interactive manner. However, the company advises that the assistant won’t access the internet and may provide outdated or inaccurate information in certain situations. An enterprise-focused version is also in development, featuring self-deployment and advanced moderation capabilities.

Mistral has successfully raised over $500 million in funding through seed and series A rounds, backed by prominent investors including Lightspeed Venture Partners and Andreessen Horowitz (a16z).