Mistral AI Launches Advanced Multilingual Model, Partners with Azure

Mistral Large, the latest and most advanced language model from Mistral AI, is now available to developers through multiple channels, including Azure. This flagship model boasts top-tier reasoning capabilities and excels in complex multilingual tasks, making it a strong contender in the world of large language models (LLMs).

Technical Breakdown: What’s New with Mistral Large

Mistral Large introduces several key advancements that set it apart from its predecessors:

• Multilingual Fluency: Natively fluent in English, French, Spanish, German, and Italian, Mistral Large demonstrates a nuanced understanding of grammar and cultural context. This makes it highly versatile for international applications.

• 32K Token Context Window: With a 32K token context window, Mistral Large can handle large documents with precision, ensuring that information is recalled accurately and contextually.

• Precise Instruction Following: The model’s ability to follow instructions precisely enables developers to implement robust moderation policies. This feature was crucial in setting up the system-level moderation for le Chat, Mistral's conversational AI.

• Native Function Calling: Mistral Large supports function calling natively, which, combined with constrained output mode (available on la Plateforme), facilitates application development and tech stack modernization at scale.

Benchmark Performance

Mistral Large has been benchmarked against leading models such as GPT-4, Claude 2, Gemini Pro 1.0, GPT 3.5, and LLaMA 2 70B on the MMLU (Measuring Massive Multitask Language Understanding) test. Here are the key findings:

• MMLU Scores: Mistral Large ranks second globally among models available via API, just behind GPT-4.

• Task Performance: The model excels in tasks requiring complex reasoning, text understanding, transformation, and code generation.

Availability Channels

Mistral AI is committed to making its models accessible through multiple platforms:

1. La Plateforme: Hosted on Mistral’s infrastructure in Europe, this platform provides developers with a secure environment to build applications using a range of Mistral models.

2. Azure: In partnership with Microsoft, Mistral Large is now available through Azure AI Studio and Azure Machine Learning. Beta users have reported significant success with the model, highlighting its performance and ease of integration.

3. Self-Deployment: For highly sensitive use cases, organizations can self-deploy Mistral models on their own infrastructure, with access to the model weights. Success stories from such deployments are available on Mistral’s website, and interested parties can contact the team for more details.

Why It Matters

The release of Mistral Large represents a significant step forward in the capabilities of multilingual LLMs. Its availability on Azure also marks a strategic partnership that could accelerate the adoption of advanced AI solutions across various industries. For developers, this means access to a powerful tool that can handle complex tasks with precision and efficiency.

Conclusion

Mistral Large is more than just an update; it’s a leap forward in the capabilities of language models. With its multilingual fluency, large context window, precise instruction following, and native function calling, it stands out as a top-tier model for both research and practical applications. The partnership with Azure further underscores Mistral AI's commitment to making cutting-edge AI accessible to a broader audience.