Meta introduced its newest and biggest open-source AI model, Llama 3.1, featuring an expanding context length and impressive benchmark performances.
The tech giant revealed that the new AI model outperformed OpenAI and other companies in terms of certain benchmarks.
(Photo : Fabrice Coffrini/AFP via Getty Images)
Read Also: Instagram Partners With COS to Study Its Effect on Teens' Mental Health
Meta Releases Llama 3.1 Models for Various Uses
In a blog post, Meta unveiled three new models under Llama 3.1. The 405B is the flagship foundation model that can be used in various cases, while the 70B is recommended for high-performance yet cost-efficient cases. The 8B is an ultra-fast model that can be run anywhere.
"In addition to having significantly better cost/performance relative to closed models, the fact that the 405B model is open will make it the best choice for fine-tuning and distilling smaller models," Meta CEO Mark Zuckerberg wrote in a blog post.
Meta claimed that Llama 3.1 outperformed OpenAI's GPT-4o and Anthropic's Claude 3.5 Sonnet on different benchmarks. The new AI model is considered more complex than the Llama 3 models that were first introduced a few months back.
Meta Continues to Expand AI Ecosystem
Zuckerberg shared that the company is currently working with different companies to continue growing its ecosystem. The Llama 3.1 models will soon be available on AWS, Azure, Google, Oracle, and more.
The CEO also defended the company's choice to develop open-source AI models. Zuckerberg stated that this type of AI model is improving faster than proprietary models.
He also argued that open sourcing does not open doors for a massive advantage. Meta will continue to become consistently competitive and efficient for every generation.
The company is actively building internal teams and is inviting developers and partners that could utilize Llama for more customer functionality.
Related Article: Meta Lifts Donald Trump's Account Restriction on Instagram, Facebook