Meta’s Llama artificial intelligence models are being used by companies including Goldman Sachs and AT&T for business functions like customer service, document review and computer code generation, the social media giant said in a statement on Thursday.

The mostly free-of-charge Llama models have been downloaded almost 350 million times since Meta began releasing them publicly last year, an increase from the 300 million downloads the company announced when it released the biggest version of its latest Llama 3 model in late July.

Usage via cloud providers like Amazon Web Services and Microsoft Azure has also increased, more than doubling between May and July this year, Meta said.

The announcement comes as Meta and other tech companies plowing billions into AI have faced questions from investors about how widespread adoption of the technology appears to be and what kind of payoff they can expect for their spending.

Meta Chief Executive Mark Zuckerberg has argued that building state-of-the-art AI models and giving them away for free best positions the company to avoid falling into a position where it is constrained by a competitor’s closed technology.

“The path for Llama to become the industry standard is by being consistently competitive, efficient, and open generation after generation,” Zuckerberg wrote last month.

Although large language models like Llama have wowed users with their ability to generate human-like prose on command, they continue to struggle with certain logical tasks and are prone to making factual errors, limiting their adoption in business contexts.

Still, Meta pointed to a handful of large enterprises dabbling with the Llama models as both evidence of their utility and as an endorsement of their competitiveness with paid alternatives, like models from industry leader OpenAI.

Other companies it said were using Llama include Japanese bank Nomura Holdings, food delivery service DoorDash and professional services provider Accenture.