Meta and Mistral Release New-Generation Generative AI Models with Enhanced Multilingual Support
On July 23, 2024, Meta announced its Llama 3.1 series of generative AI models. The following day, Mistral AI followed suit by releasing the Mistral Large 2 model. These new models represent significant advancements in multilingual support and efficiency, poised to further the development of AI applications in languages like Spanish.
The Release of Meta Llama 3.1 Series
The Llama 3.1 series includes 8B, 70B, and 405B parameter models[\/color], with the 405B model being the first open-source AI model considered to have reached the frontier level of performance. Its context length has been expanded to 128K tokens, and it supports 8 languages: English, French, German, Italian, Portuguese, Spanish, Hindi, and Thai. It demonstrates outstanding performance in general knowledge, mathematics, tool use, and multilingual tasks, comparable to leading closed-source models. The models are available for download on llama.meta.com and Hugging Face, with support from multiple cloud platforms.
Advanced Features of Mistral Large 2
Mistral Large 2 has 123B parameters and also features a 128K context window. This model particularly enhances code generation, mathematics, and reasoning capabilities. Its training data includes a large amount of code, supporting over 80 programming languages. In terms of multilingual capabilities, it supports various languages, including Spanish. Benchmark tests show its performance is comparable to GPT-4o, Claude 3 Opus, and Llama 3.1 405B.
Model Availability and Impact
The two companies provide access to their models in different ways. Meta emphasizes open source, while Mistral offers its models via API and licensed weights. These releases lower the barrier to entry for generative AI development and foster innovation within the global developer community, especially benefiting the efficiency and accessibility of AI applications in non-English speaking regions.
The Continuous Evolution of Generative AI Technology
This latest update from Meta and Mistral reflects the industry’s trend towards larger, more intelligent, and more inclusive models. The enhancement of multilingual capabilities will help more users and developers leverage AI technology to solve real-world problems, promoting the global adoption of the technology.
