Mistral Large Launches as AI Startup Secures Major Microsoft Partnership
French AI startup Mistral unveils Mistral Large, a powerful multilingual model that ranks just behind GPT-4 on key benchmarks, alongside a new strategic partnership with Microsoft.
Mistral releases Mistral Large, its newest and most capable enterprise text generation model, alongside a strategic partnership with Microsoft. This collaboration gives the one-year-old AI startup a massive new distribution channel through Azure AI and brings in $16 million in fresh capital. The new model is available immediately through the Mistral API and Azure.
The new model handles complex multilingual reasoning tasks with native fluency in English, French, Spanish, German, and Italian. Mistral Large features a 32K token context window for processing large documents and includes precise instruction-following capabilities that allow developers to design custom moderation policies. The startup also launches an updated version of its smaller Mistral Small model and a new chat app for business teams.
According to the MMLU benchmark, Mistral Large achieves an accuracy of 81.2%, placing it second only to GPT-4 among generally available API models. This score easily surpasses competing models like Gemini Pro 1.0 and Llama 2 70B. While it remains to be seen how it compares to newer massive-context models in real-world applications, the benchmark results establish Mistral as a strong competitor in the enterprise AI space.