Open-source AI is rapidly gaining ground against its closed-source counterparts. Recently, Meta, a key player in the open model space, reported a significant surge in adoption of its Llama family of models, especially following the recent release of Llama 3.1.
According to Meta, Llama model downloads on Hugging Face are approaching 350 million, representing more than a tenfold increase from last year. The company highlighted how major enterprises, including Zoom, Spotify, Infosys, AT&T, and Goldman Sachs, are leveraging these models for both internal and external applications.
This update illustrates that open-source AI, previously seen as lagging, is now matching closed-source offerings in performance and gaining substantial traction within enterprises, fueling real-world applications. It raises questions about the dominance of OpenAI, which has faced criticism for not delivering advanced AI products beyond mere announcements.
Meta’s Open-Source Success
While OpenAI initially led the generative AI landscape, Meta quickly entered the field with its Llama model, which was launched just three months after the ChatGPT model. However, Meta opted for an open-source approach, making the Llama ecosystem—including Llama 2, Llama 3, and the latest Llama 3.1—available through Hugging Face and various cloud partners.
“Making our Llama models openly accessible has fostered a vibrant, diverse AI ecosystem where developers enjoy greater choice and capability. The innovation has been swift and widespread, with start-ups and enterprises alike pushing boundaries using Llama,” the company stated in a blog post.
French startup Mistral has adopted a similar open-source strategy, providing developers with robust models to create derivatives that match or even exceed the performance of closed models on select metrics (e.g., FinGPT, BioBert, Defog SQLCoder, and Phind).
Since the launch of Llama 2 in July 2022, the downloads of Meta’s models have skyrocketed, reaching close to 350 million. In July alone, the company recorded over 20 million downloads, demonstrating strong momentum following the release of Llama 3.1 405B.
“We're witnessing a growing preference among developers for Llama, with indicators suggesting continued growth. A survey from Artificial Analysis indicates that Llama ranks as the second-most considered model and leads in the open-source space,” Meta added.
Enterprise Adoption
Several major enterprises, including AT&T, DoorDash, Goldman Sachs, Niantic, and Shopify, are utilizing the Llama family of models for various applications. Additionally, enterprise developers are accessing Llama through Meta’s extensive network of cloud and infrastructure providers, such as AWS, Microsoft Azure, Google Cloud, Groq, Nvidia, Databricks, and Snowflake. While the company has not disclosed specific numbers by partner, it confirmed that Llama's monthly input/output token usage surged tenfold from January to July 2024 across select cloud providers.
The statistics indicate consistent growth in Llama's hosted usage, peaking with the release of Llama 3.1 in July. It's important to note that these figures reflect only some of Meta's largest cloud partners, meaning actual usage numbers could be higher.
Trends in Open-Source AI
Llama's adoption by leading enterprises like AT&T and Spotify signifies that open-source AI is rapidly catching up. Improvements in performance and the long-term cost advantages of open-source solutions are pivotal drivers of this trend. As developments continue, it’s likely that open-source AI will increasingly challenge closed models, prompting companies in that space to innovate more and reduce costs.
The effects of the open-source movement are already visible. OpenAI, the pioneer in generative AI, has significantly slashed the prices of its existing models, including GPT-4o. However, in terms of product innovation, OpenAI appears to be trailing behind, with many of its recently announced offerings, like Sora and SearchGPT, either unreleased or available only to a select few users.
Ranking
