Mistral AI Announces "Mistral Large 2" — Claiming Parity With GPT-4o and Llama 3 405B
On July 24, 2024, Mistral AI announced the second generation of its flagship model, “Mistral Large 2.” The company described it as a “significant improvement” over the original Mistral Large from that February, claiming performance rivaling then-state-of-the-art models such as GPT-4o, Claude 3 Opus, and Llama 3 405B, while keeping a size that could run on a single node.
Details
- Size and design: 123 billion parameters, designed with single-node inference and long-context usage in mind. Context window of 128K
- Multilingual and coding support: Supports numerous languages including French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, and Korean, along with more than 80 programming languages including Python, Java, C, C++, JavaScript, and Bash
- Benchmarks: Recorded 84.0% accuracy on MMLU, pushing forward the performance-cost Pareto frontier for open models
- License: Released under the Mistral Research License, which permits research and non-commercial use
- Distribution partners: Available via La Plateforme as well as a broad range of platforms including Google Cloud Vertex AI, Amazon Bedrock, Azure AI Studio, and IBM watsonx
What happened next
Mistral Large 2 became a representative example of Mistral AI’s ability to “match performance despite a smaller scale.” That September, the company launched its first multimodal model, Pixtral 12B, accelerating the expansion of its model lineup.