Mistral AI has unveiled its latest groundbreaking creation, Mistral Large 2. This advanced language model represents a significant leap forward in AI capabilities, outperforming its predecessor and many competitors in several key areas. Renowned for its strong performance in code generation, Mistral Large 2 further excels in this domain, surpassing even Llama 3.1 405B, a model released just days prior. The new model also demonstrates exceptional proficiency in mathematics, ranking second only to GPT-4o on the MATH benchmark.
Mistral Large 2 boasts a substantial 123 billion parameters and a context window of 128,000 tokens, enabling it to process and generate highly complex and informative text. The model’s multilingual abilities have also seen remarkable improvement, surpassing Llama 3.1 70B base by an average of 6.3% across nine languages. Despite its size, Mistral AI has optimized the model for single-node inference, ensuring efficient performance for long-context applications.
Beyond its impressive performance metrics, Mistral Large 2 showcases enhanced reasoning capabilities and reduced hallucinations. The company’s focus on refining these aspects has resulted in a model that can tackle intricate problems and generate more reliable outputs.
Mistral AI is committed to making this powerful technology accessible to a wide audience. Mistral Large 2 is available on the company’s platform, la Plateforme, and the model’s weights have been released on Hugging Face for research purposes. Furthermore, Mistral AI has expanded its partnership with Google Cloud Platform, integrating Mistral Large 2 into Vertex AI via a Managed API.
This latest achievement solidifies Mistral AI’s position as a leader in the AI landscape, demonstrating its dedication to pushing the boundaries of language model capabilities. With Mistral Large 2, the company is poised to revolutionize various industries, from content creation and education to scientific research and software development.
