Mistral Large 2 Outperforms Llama 3.1 in Coding and Math Benchmarks
Mistral AI releases Mistral Large 2, a 123-billion-parameter model that surpasses Meta's Llama 3.1 in code generation and mathematical reasoning while supporting over 80 programming languages.
Mistral AI unveils Mistral Large 2, a powerful new flagship model that directly challenges Meta's recently released Llama 3.1. The new model features 123 billion parameters and demonstrates superior performance in code generation and mathematical reasoning compared to its competitors.
The model supports over 80 coding languages, including Python, Java, and C++, alongside dozens of human languages like French, German, Spanish, and Chinese. It also offers a 128k context window and advanced function-calling capabilities, making it highly effective for single-node inference and long-context applications.
Mistral Large 2 achieves an impressive 84.0% accuracy on the MMLU benchmark, positioning itself as a highly competitive option against leading models like GPT-4o. The company releases this model under the Mistral Research License, which permits researchers and non-commercial users to freely access and evaluate the technology.