Meta Platforms has released its latest version of its Llama 3 AI models, which can converse in eight languages, write high-quality computer code, and solve complex math problems, Reuters reports.
The new Llama 3 model has 405 billion parameters, surpassing previous versions but still smaller than competitors like OpenAI’s GPT-4 and Amazon’s upcoming 2 trillion parameter model. The Llama 3 models nearly match leading models like GPT-4o and Claude 3.5 Sonnet in key benchmarks. Meta CEO Mark Zuckerberg expects future Llama models to surpass competitors by next year, with Meta’s AI chatbot expected to become the most popular AI assistant by the end of 2024.
Meta also updated its smaller Llama 3 models, which now feature an expanded “context window” for better performance on tasks like generating computer code. The Llama 3 models nearly match leading models like GPT-4o and Claude 3.5 Sonnet in key benchmarks. Meta offers these models mostly free to developers, aiming to foster innovation and engagement on its platforms, despite some investor concerns over costs.
