Meta Platforms announced the launch of the largest version of its mostly free Llama 3 artificial intelligence models, showcasing multilingual capabilities and strong performance metrics that rival paid models like those from OpenAI.
The Llama 3 model can communicate in eight languages, generate higher-quality computer code, and solve more complex math problems than previous versions, according to a blog post and research paper from the Facebook parent company.
With 405 billion parameters, the model’s size surpasses its predecessor released last year but remains smaller than leading models from competitors. For comparison, OpenAI’s GPT-4 reportedly has one trillion parameters, and Amazon is developing a model with 2 trillion parameters.
This release is part of a broader competition among tech companies to demonstrate the value of their advanced, resource-intensive large language models. These models aim to improve areas like advanced reasoning, justifying the significant investments made in them.
Alongside the flagship 405 billion parameter model, Meta is also updating its lighter-weight 8 billion and 70 billion parameter Llama 3 models, initially introduced in the spring. All three models support multiple languages and can manage larger user requests due to an expanded “context window.” Ahmad Al-Dahle, Meta’s head of generative AI, stated that this feature enhances the models’ ability to generate computer code, as it provides a longer “memory” for handling multi-step requests.
Meta offers its Llama models mostly free for developers, a strategy that CEO Mark Zuckerberg believes will lead to innovative products and increased engagement on Meta’s social platforms. However, some investors are concerned about the costs involved.
The company could also benefit if developers choose its free models over paid alternatives, potentially disrupting the business models of its competitors. Meta highlighted improvements in key math and knowledge tests, which could make their models more attractive to developers.
