Anthropic has been gaining on arch rival OpenAI in recent months, according to data from a popular developer tool offered by startup Vercel.

The AI SDK is a toolkit that helps developers build AI-powered applications. It provides code libraries that developers install to get their apps to communicate with different AI models. Vercel publishes data showing how many times these code packages are downloaded each week.

This is a useful window into how popular the top AI models are with developers, a key constituency of early tech adopters.

The data shows that OpenAI remains in the lead. However, Anthropic has been growing faster lately in terms of developer mindshare.

In the week of August 18, the OpenAI code library was downloaded almost 229,000 times by developers. The equivalent Anthropic library was downloaded almost 38,000 times, according to Vercel data.

In late October, OpenAI downloads had remained relatively unchanged, while Anthropic downloads were much higher: about 244,000 for OpenAI and 100,000 for Anthropic.


Data from Vercel's AI SDK toolkit

Data from Vercel’s AI SDK toolkit

Vercel



A top executive at Vercel attributed the recent adoption trends partly to a new AI model that Anthropic released in the summer.

“Anthropic’s recent gains can be attributed to their Claude 3.5 Sonnet model, which, among many tasks, excels at understanding and generating code,” said Jared Palmer, VP of AI at Vercel. “Its combination of speed, cost, and output quality has won over the developer community for the time being.”

Claude 3.5 Sonnet came out in June, and it has been particularly useful for tasks that developers need to do regularly, such as understanding and generating software code.

Still, OpenAI has maintained its lead among developers, according to the Vercel data. That’s primarily due to its first-mover advantage and other factors such as cheap pricing of its GPT-4o-mini AI model and strong developer tooling and documentation.

There’s also pretty widespread adoption of their API, or application programming interface. This is a common way for different software applications to communicate and share data.

It’s worth noting, though, that OpenAI’s lead is slightly inflated by the download data from Vercel’s AI SDK service. That’s because many AI model providers have adopted OpenAI’s API specification as a de facto standard. That allows developers to use the same OpenAI code library package but point their apps to different models.

It’s hard to quantify how much this exaggerates OpenAI’s lead, but it also supports the general conclusion that Anthropic has been gaining on its main rival in recent months.

An OpenAI spokesperson said usage of the startup’s API has doubled since it released its GPT-4o mini AI model in July. OpenAI has also tripled the number of active apps running in production on its platform in the past year.

The OpenAI spokesperson shared a critique of Vercel’s data, saying downloads of these code libraries are not technically an accurate representation of developer audience since it doesn’t reflect actual usage or spending.

Anthropic had no problem with the data and what it shows. The company said its Claude 3.5 Sonnet model helps developers handle complex software architecture and multi-step processes. It also noted that enterprises are increasingly adopting Claude for business-critical functions where accuracy is key, such as software development and data analysis.

“Our new Claude 3.5 Sonnet model is the ideal choice for developers to build AI-powered applications with state-of-the-art performance across the board, including coding, complex financial analysis, and live chat support,” Dianne Penn, product lead at Anthropic, told Business Insider.

Penn cited DoorDash, Sourcegraph, and GitLab as companies that are using Claude to reduce development time, increase code acceptance rates, and build AI features.