Meta announces open-source AI model to rival OpenAI
Written by Rebecca Uffindell Thu 25 Jul 2024
Meta has claimed its new artificial intelligence (AI) model is the first open-source system that can compete with those from OpenAI and Anthropic.
Meta said the new model, named Llama 3.1 405B, is competitive with leading foundation models across a ‘range of tasks’, including OpenAI’s GPT-4, GPT-4o, and Anthropic’s Claude 3.5 Sonnet.
If accurate, this would be the first time one of the world’s most powerful AI models is available without intermediary charging for access or controlling its use.
“Until today, open source large language models have mostly trailed behind their closed counterparts when it comes to capabilities and performance. Now, we are ushering in a new era with open source leading the way,” said Meta in a blog post.
The New Llama Model
Meta is releasing the Llama 3.1 405B model, marking the debut of the first frontier-level open-source AI model. Alongside it, upgraded versions of the Llama 3.1 70B and 8B models are being introduced. The 405B model offers better cost and performance compared to closed models. Meta added the new model is ideal for fine-tuning and distilling smaller models.
With the 405B model, Meta aims to accelerate innovation by opening new avenues for growth and exploration. This latest generation of Llama is expected to drive advancements in applications and modelling, like synthetic data generation for enhancing smaller models and model distillation at an unprecedented scale in open source.
The updated 8B and 70B models feature multilingual capabilities, a longer context length of 128K, advanced tool use, and enhanced reasoning abilities. These improvements support use cases like long-form text summarisation, multilingual conversational agents, and coding assistants. Meta has also revised its licensing terms, allowing developers to use outputs from Llama models, including the 405B, to refine other models.
In line with its commitment to open source, Meta is making these models available for download on llama.meta.com and Hugging Face, and they are ready for immediate development on a wide range of partner platforms.
Meta Commits to Open Source
In a blog post, Founder and CEO of Meta, Mark Zuckerberg, said the company champions open source for several reasons.
Historically, the shift from closed Unix systems to open source Linux revolutionised technology by offering greater flexibility, affordability, and innovation. Meta anticipates a similar shift in AI, where open source models like Llama will drive significant advancements, cost efficiency, and industry leadership.
For Meta, open source aligns with its values by avoiding vendor lock-in and fostering a broad ecosystem of tools and partnerships. This approach, which has been successful with projects like PyTorch and React, supports ongoing innovation and ensures access to cutting-edge technology without constraints.
Globally, open source AI democratises technology, offering greater transparency and security while preventing power from being concentrated in a few companies. This fosters a more inclusive and competitive environment, ensuring AI’s benefits are widely accessible and enhancing global technological progress and safety.
“The bottom line is that open source AI represents the world’s best shot at harnessing this technology to create the greatest economic opportunity and security for everyone,” said Zuckerberg.
Written by Rebecca Uffindell Thu 25 Jul 2024
Most Viewed News
Fri 2 Oct 2026
AI skills gap grows despite 64% of workers using AI