Mistral Large 2 Released

Mistral AI has introduced Mistral Large 2, a model with a 128k context window and 123 billion parameters, designed for high-performance AI applications. It supports numerous languages and coding languages, aiming for efficient single-node inference.

• Performance Metrics: The model achieves an 84.0% accuracy on MMLU, outperforming previous versions and competing with top models like GPT-4o and Claude 3 Opus. Enhanced reasoning capabilities reduce the model's tendency to generate inaccurate information.
• Improved Instruction Following: Mistral Large 2 shows significant improvements in following instructions and engaging in multi-turn conversations, with a focus on concise responses for business applications.
• Multilingual Proficiency: The model excels in various languages, making it suitable for multilingual business use cases. It has been trained on a diverse dataset to enhance its language capabilities.
• Function Calling Enhancements: Mistral Large 2 features improved function calling and retrieval skills, making it effective for complex business applications.
• Availability: Users can access Mistral Large 2 on la Plateforme and through cloud services like Google Cloud, Azure, Amazon Bedrock, and IBM watsonx.ai, with fine-tuning capabilities now expanded for several models