Mistral announces Large 2: flagship LLM with 123 billion parameters

Large 2 - the leading large language model (LLM) with significantly higher code generation, mathematical computation, and reasoning capabilities. Mistral has also added improved multi-language support and a range of advanced functions with the Large 2.

If you don't know, a large language model is a language model with general capabilities for language generation and other natural language processing tasks. LLM achieves this ability by learning statistical relationships from texts during highly computationally complex self-supervised and semi-supervised training.

The Mistral Large 2 model has 123 billion parameters, allowing it to run on a single H100 node with high throughput. This LLM has comprehensive support for French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese and Korean. In terms of coding, Large 2 supports over 80 different programming languages, including Python, Java, C, C++, JavaScript and Bash…

Large 2 is currently available for open access, but is only made freely available by Mistral for research and non-commercial purposes. For commercial use, users need a specialized use license.

With 123 billion parameters (123B), Mistral Large 2's performance is comparable to GPT-4o, OpenAI's Claude Opus 3, and Meta's recently released Llama 3.1 405B in terms of encoding capabilities. On the Wild Bench, Arena Hard and MT Bench ratings, Large 2 outperformed Llama 3.1 405B and Claude 3 Opus. On the popular MMLU benchmark, this new model performs better than the Llama 3.1 70B and is comparable to the Llama 3.1 405B.

From a developers perspective, Mistral Large 2 now has improved function calling and retrieval skills. The model can now execute both parallel and sequential function calls, allowing developers to build complex business AI applications.

With the release of Large 2, Mistral's LLM ecosystem is now relatively diverse, including Mistral Nemo, Mistral Large, and two specialized models: Codestral and Embed. Mistral will discontinue the Apache models (Mistral 7B, Mistral 8x7B and 8x22B, Codestral Mamba, Mathstral) in the future.

Microsoft and Mistral have a partnership to integrate Mistral models on Azure. Today, Mistral is expanding its partnership with Google to bring its products to Google Cloud.

The consecutive releases of Mistral Large 2 and Llama 3.1 mark a major milestone for the open AI ecosystem, providing two powerful GPT-4 level models for research and development. This rapid progress drives growing momentum towards a more open and collaborative AI ecosystem.

Isabella Humphrey

Update 25 July 2024