Mistral Large 2 AI Model Boosts Code Generation and Reasoning

Mistral AI has announced the release of Mistral Large 2, a significant upgrade to its flagship model. This new generation boasts enhanced capabilities in code generation, mathematics, and reasoning, as well as stronger multilingual support and advanced function calling features.

With a 128k context window, Mistral Large 2 supports dozens of languages, including French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, and Korean, along with over 80 coding languages.

The model’s size of 123 billion parameters allows it to run at large throughput on a single node, making it an attractive option for businesses and researchers alike. Mistral AI has released Mistral Large 2 under the Mistral Research License, allowing usage and modification for research and non-commercial purposes.

Introducing Mistral Large 2: A New Generation of AI Capabilities

Mistral AI has announced the release of Mistral Large 2, a significant upgrade to their flagship model. This new generation boasts enhanced capabilities in code generation, mathematics, and reasoning, as well as improved multilingual support and advanced function calling features.

Enhanced Performance and Cost Efficiency

Mistral Large 2 is designed for single-node inference with long-context applications in mind, featuring a 128k context window and an impressive 123 billion parameters. This allows it to run at large throughput on a single node while maintaining cost efficiency. The model has been fine-tuned to minimize “hallucinations” or generation of plausible-sounding but factually incorrect information, ensuring reliable and accurate outputs.

Code Generation and Reasoning Capabilities

Mistral Large 2 has been trained on a vast proportion of code, vastly outperforming its predecessor and performing on par with leading models such as GPT-4o, Claude 3 Opus, and Llama 3 405B. The model’s reasoning capabilities have also been significantly enhanced, demonstrating improved performance on popular mathematical benchmarks.

Instruction Following and Alignment

The new Mistral Large 2 excels in following precise instructions and handling long multi-turn conversations. It has shown impressive performance on MT-Bench, Wild Bench, and Arena Hard benchmarks, making it an ideal choice for business applications that require concise and accurate responses.

Language Diversity and Multilingual Support

Mistral Large 2 has been trained on a large proportion of multilingual data, exceling in languages such as English, French, German, Spanish, Italian, Portuguese, Dutch, Russian, Chinese, Japanese, Korean, Arabic, and Hindi. The model’s performance on the multilingual MMLU benchmark is impressive, outperforming its predecessor and other leading models.

Tool Use and Function Calling

Mistral Large 2 is equipped with enhanced function calling and retrieval skills, proficiently executing both parallel and sequential function calls. This enables it to serve as the power engine of complex business applications.

Availability and Access

Mistral Large 2 is available for use on la Plateforme under the name mistral-large-2407, and can be tested on le Chat. The model’s weights are also hosted on HuggingFace. Mistral AI has consolidated its offering on la Plateforme around two general-purpose models, Mistral Nemo and Mistral Large, and two specialist models, Codestral and Embed.

Additionally, Mistral AI is extending fine-tuning capabilities on la Plateforme for Mistral Large, Mistral Nemo, and Codestral. The company has also partnered with leading cloud service providers to bring the new Mistral Large 2 to a global audience, including Google Cloud Platform, Azure AI Studio, Amazon Bedrock, and IBM watsonx.ai.