Our open source Llama models are helping to spur economic growth in the US

Meta’s Llama open source AI models help to unlock innovation and competition, letting people build exciting new tools that positively impact the US economy and our daily lives. Llama, which has been downloaded more than a billion times, is revolutionizing the way American businesses operate, enhancing productivity and efficiency, creating new opportunities for growth, and unlocking a new wave of entrepreneurship and American homegrown innovation.

Accessing AI tools should be easy, and unlike closed models that come with high costs and restrictive access, Llama is free, available to all, and gives developers control to run their models wherever they want without needing to build from scratch because it’s an open model. That means startups, small businesses, and other innovators without deep pockets can use Llama to build their business. We’re committed to developing open source AI systems like Llama, as open source is a critical piece of ensuring America’s geopolitical leadership by leveling the playing field so more people and American businesses can access AI to compete in the global economy.

Open source isn’t altruism—it’s good for Meta, too. When other companies and developers test and build on top of AI, we get to learn from their innovations, which in turn lets us improve our own models. The only way for Llama to become an industry standard is if it remains consistently competitive, efficient, and open generation after generation.

We’ve shared how Llama’s impact has already been felt by businesses and entrepreneurs. And today, we’re looking at how Llama is helping to spur economic growth in the US.

Based in Tulsa, Okla., WriteSea uses Llama—specifically our lightweight 3B Instruct model—to build Job Search Genius, an AI career coach designed to improve the experience of looking for work and help candidates stand out.

The average job search can take five to six months, and WriteSea is fully focused on helping job seekers land their next job 30% to 50% faster for up to one-sixth of the cost of traditional job search strategies. While people normally see a 1% response rate for cold outreach applications, those using Job Search Genius experience a 2.32% response rate—that means they have more than double the chance of hearing back from a recruiter when using a resume created using WriteSea’s tools.

Although WriteSea started out using closed source models, the team decided to explore open source models with Llama and quickly realized three primary benefits: cost effectiveness, data security, and a vibrant developer community.

“Cost matters,” says WriteSea Co-Founder & CEO Brandon Mitchell. “Instead of paying for these super scaled API calls for a closed source model, you can control your cost when you're building on top of Llama. It’s a fixed cost because you’re not paying per API call. With open source, you can scale the right way.”

Because resumes include a lot of personally identifiable information (PII), data security is paramount—and Llama met these needs. “Because we can deploy and fine-tune everything locally on our own servers, we have full security of our data,” says Mitchell. “We have 100% certainty that it’s not being accessed.”

Finally, WriteSea benefits from the large—and growing—community of fellow Llama developers. “Just tapping into the developer community—being able to quickly figure out solutions to problems, talking to other developers, and seeing what's out there—I think that’s huge,” Mitchell explains. “You can’t shine a light brightly enough on that.”

Outside of her day job as a machine learning engineer, Srimoyee Mukhopadhyay spends her downtime building a cultural tourism app using Llama in Austin, Texas.

A UNESCO City of Media Arts, Austin is a great place for people to explore local history and broaden their understanding of not just Texan history, but American history, too. In addition to live music performances, the city is home to beautiful murals, statues, and other works of art that aren’t well documented.

“The outside walls of local cafés have really beautiful murals,” explains Mukhopadhyay, who won the Local Impact Prize at our 2024 Austin Llama Impact Hackathon. “And they were built like 40 years ago. They represent an important part of the evolving culture in Austin. Thanks to our app and Llama’s new vision model, you can take a photo and the Llama model gives you the history behind that particular thing you captured, including how it relates to the culture and history of Austin. I use Llama to transform cities into living museums, revealing their best kept secrets: murals, street art, and lost histories.”

Given that Mukhopadhyay’s app needs to run on a phone while the user is on the go, it was critical to find a lightweight model that could run locally rather than on the cloud.

“Llama is great,” says Mukhopadhyay. “With the latest updates, it can run on-device, which means the app doesn’t need internet connectivity, and that really helps for walking tours where internet may not be always available..”

And because Mukhopadhyay’s app redirects foot traffic to areas that aren’t typically recommended as tourist attractions, local businesses can reap the benefits.

“If you check out a beautiful mural on the side of a taco joint, you’re more likely to grab a taco,” Mukhopadhyay says. “If you’re learning about the history of a mural outside a café, you might stop in for a coffee. The app redirects traffic everywhere and helps attract more tourists to lesser-known areas, which helps bolster the local economy.”

Based in Austin Texas, Fynopsis leverages Llama to facilitate more efficient and accurate deals in the mergers and acquisitions (M&A) space—a key tool in helping smaller and lower-middle market businesses get ahead. They’re also targeting private equity (PE) diligence. Through Capital Factory’s Longhorn Startup program, the team has been connecting with local CEOs, including those from PE firms, to refine its solutions based on their insights.

“M&A analysts use something called virtual data rooms, which you can think of as a very confidential drive or folder where both parties can exchange company documents and information securely,” says Fynopsis CEO & Co-Founder William Zhang. “But a lot of the incumbent providers are very outdated—they’re really lacking AI capabilities, and they’re not open source. We believe that open source is huge in the business space because it brings transparency and an increase in security to all the products that people use. And the smaller 8B Llama model is an absolute powerhouse—lightweight, cost-effective, and fast—making it perfect for our front-end user experiences.”

Thanks to Llama, Fynopsis is attempting to streamline M&A workflows and cut the amount of time it takes for due diligence in half so more deals can be closed faster.

“Virtual data rooms can be incredibly expensive—up to $80,000 in more expensive cases,” explains Zhang. “That’s a lot of money. And for small and medium-sized businesses with more constrained budgets and smaller teams, it’s not really an option. They’re often forced to use more rudimentary methods to share confidential data, which isn’t practical. We want to help them gain leverage in this space and take control of their work using AI.”

Fynopsis initially tested closed source models, but found the lack of transparency to be a roadblock when it came to fine-tuning the models.

“In our business, we have to fine-tune the models for very specific use cases, and we don’t have any room for error,” Zhang says. “If you get a number or the analysis wrong, that could cost the entire deal. With Llama, we had the transparency that we needed. And because Llama is open source, it brings with it a lot of innovation. We looked at Groq, which uses a Llama license and its architecture to speed up AI inference by about 10x. Because it’s open source, we’re able to leverage all the innovation that comes with Llama. It’s the full package. While we’re still using Groq, we’ve reduced our dependency on it and moved much of our inference to serverless options like Modal and Ollama to host our fine-tuned Llama models. Things are evolving rapidly!”

Long-term, Fynopsis is evolving into a launchpad for AI agents specialized in due diligence. According to Zhang, “Our hybrid RAG architecture, enhanced by lightweight Llama models, provides a cutting-edge foundation for iterative agent development.”