Le mistral current models explained

Here's a comprehensive breakdown of Mistral's key AI models:

Premier Models

Mistral Large 2

• Most advanced model with 123 billion parameters
• 128k context window
• Excels in reasoning, code generation, and mathematics
• Supports dozens of languages and 80+ coding languages
• Released July 2024[3]
• Available under Mistral Research License for non-commercial use[3]

Codestral

• 22B parameter model specialized for code generation
• Open-weight model for research and non-commercial use
• Supports 80+ programming languages
• Features 32k context window
• Focuses on code completion and generation tasks[4]

Mistral Nemo

• 12B parameter model with 128k context window
• Strong reasoning and instruction-following capabilities
• Designed for enterprise applications
• Outperforms larger models like Llama-2-13B[2]

Multimodal Model

Pixtral 12B

• First open-source multimodal model by Mistral
• Processes both images and text
• 40 layers architecture
• Supports images up to 1024×1024 resolution
• Released September 2024[5]

Open Source Models

Mixtral 8x7B

• Outperforms Llama 2 70B on most benchmarks
• 6x faster inference than comparable models
• Strong multilingual capabilities[2]

Mixtral 8x22B

• Most performant open model
• Handles multiple languages
• Strong code-related performance[2]

Performance Comparison

| Feature | Speed (tokens/s) | Latency (seconds) | |