mistral-ai

Introducing Mistral Large 2: The New Flagship Model from Mistral AI

Tal Peretz

— 1 min read
Introducing Mistral Large 2: The New Flagship Model from Mistral AI

On July 24, 2024, Mistral AI unveiled its latest flagship model, Mistral Large 2. This advanced language model is now available on multiple cloud platforms, including Google Cloud's Vertex AI, Azure AI Studio, Amazon Bedrock, and IBM Watsonx.ai.

Capabilities and Performance

Mistral Large 2 boasts a 128,000-token context window and supports dozens of languages alongside over 80 coding languages. Optimized for single-node inference and long-context applications, this model features 123 billion parameters. It outperforms its predecessor and rivals other leading models like GPT-4o, Claude 3 Opus, and Llama 3 405B. Key enhancements include improved reasoning capabilities, better instruction-following, and enhanced conversational handling.

Use Cases

Mistral Large 2 is ideal for:

  • Complex multilingual tasks
  • Text understanding and transformation
  • Code generation
  • Mathematical reasoning

Integration with Vertex AI

Available as a fully managed service on Vertex AI, users can access Mistral Large 2 via API without the need to manage infrastructure. The model is offered under a pay-as-you-go pricing model, allowing for flexible usage. Vertex AI's comprehensive tools support experimentation, fine-tuning, and deployment.

Additional Models

Mistral AI also offers:

  • Mistral Nemo: A cost-efficient model for low-latency workloads and simple tasks like text generation, classification, and customer support.
  • Codestral: An open-weight generative AI model specifically designed for code generation, available exclusively on Google Cloud's Vertex AI.

Usage and Access

To use Mistral Large 2 on Vertex AI, you need to enable the Vertex AI API, ensure billing is enabled for your project, and access the model through the Vertex AI Model Garden. For API calls, use the model name mistral-large@2407 for Mistral Large (previous version) and check for the latest version suffix for Mistral Large 2.

Key Features and Improvements

  • Multilingual Support: Supports dozens of languages and over 80 coding languages.
  • Reasoning and Instruction Following: Enhanced reasoning capabilities and improved instruction-following and conversational handling.
  • Cost Efficiency: Designed to deliver high performance with fewer compute resources, making it cost-efficient.

For the most current and detailed information, refer to the official announcements and documentation from Mistral AI and Google Cloud.

Read more