Mistral Large 2 Overview

Mistral Large 2 is a flagship large language model released by French AI company Mistral AI on July 24, 2024. It features 123 billion parameters and a 128,000-token context window. The model achieves 84.0% accuracy on MMLU, with code generation and mathematical reasoning capabilities approaching those of GPT-4o and Llama 3.1 405B. It supports dozens of natural languages (including English, French, German, Chinese, Japanese, Korean, etc.) and over 80 programming languages. A key focus during training was minimizing hallucinations—the model is trained to acknowledge when it doesn’t know the answer. It also features native function calling and JSON output for seamless integration with external systems and APIs. Mistral Large 2 is released under the Mistral Research License for non-commercial research use.

Mistral Large 2 Pricing

Plan	Price	Description
Mistral Official API	Input: $2 / Output: $6 per 1M tokens	Supports a 128K-token context window. Mistral Large 2 has been retired, and Mistral recommends migrating to the newer Mistral Large 3, priced at approximately Input: $0.50 / Output: $1.50 per 1M tokens.
API via Together AI	Input & Output: $2.50 per 1M tokens	Access Mistral Large 2 through the Together AI platform with a unified token-based pricing model. Pricing follows Together AI’s standard rates.
Le Chat Free / Subscription	Free / ~$14.99 per month	Mistral’s official AI chat platform. The free plan provides basic access with usage and rate limits, while the subscription plan offers higher usage capacity and additional features.

Mistral Large 2 Key Features

1. Powerful Reasoning: 84.0% MMLU accuracy, excels in math and logical reasoning
2. Code Generation Expert: Trained on 80+ programming languages including Python, Java, C, C++, JavaScript, Swift, Fortran, and more
3. Multilingual Support: Covers dozens of languages including English, French, German, Spanish, Chinese, Japanese, Korean, Arabic, Hindi, and more
4. 128K Long Context: Can process approximately 300 pages of text in a single pass
5. Agentic Capabilities: Native function calling and JSON output for seamless interaction with external systems, APIs, and tools
6. Low Hallucination Design: Trained to minimize hallucinations and acknowledge when it doesn't know the answer
7. Efficient Single-Node Deployment: 123B parameters run at high throughput on a single H100 node

Summary

1. AI Researchers: Those needing a powerful open-source research model for experimentation and fine-tuning
2.Software Developers: Engineers requiring code generation, code completion, and programming assistance
3.Enterprise Users: Organizations building complex business AI applications, agentic workflows, and function calling systems
4.Multilingual Application Developers: Teams building cross-language services and applications for multilingual users
5.RAG System Developers: Those building applications requiring long-context document processing, summarization, and retrieval-augmented generation

Model	Context	Pricing	API	Released	Global Heat
Mistral Large 2	128K	Freemium	YES	2024-07	84/100
Mistral Medium 3.5	262K	—	NO	2026-04	71/100

Mistral Large 2

Mistral Large 2 Overview

Mistral Large 2 Pricing

Mistral Large 2 Key Features

Summary

Comments (0)

Leave a comment Cancel reply

Mistral AI Model Comparison

Similar Models