Anthropic Makes Play for Business Customers

The offering, dubbed the Claude Enterprise plan, and which includes security controls and integrations with other enterprise tooling, marks the latest effort targeting company tech leaders eager to put AI models to the test, but concerned about the safety of corporate data.

“We heard from lots of folks, like: oh, we want this, we just can’t adopt it until you make sure that these pieces are all done,” Anthropic Chief Product Officer Mike Krieger said. The Instagram co-founder joined Anthropic in May. 

The Claude Enterprise plan includes the basic features required for a scaled-out enterprise deployment—such as single sign-on and identity management—as well as additional features, including a larger context window and an integration with coding software GitHub. 

It will be available starting Wednesday, after a beta test in August. It defaults to run on Claude 3.5 Sonnet, although Anthropic’s other models, Haiku and Opus, are also available on it. 

Previously, its Claude large language models were available to individuals as a free online version or a Pro version for $20 a month per user, and more recently as a “Teams” version— which allows collaboration across multiple users—for $25 a month per user. 

Anthropic said the Claude Enterprise plan pricing will vary based on seats and uses, but overall it will be more expensive than its Teams offering. 

With the announcement, Anthropic joins an increasingly competitive (and convoluted) field of AI startups and established tech players selling AI to the enterprise.

Competing model makers like OpenAI and Google, which has agreed to invest up to $2 billion in Anthropic, have a head start in their push into enterprise settings. OpenAI said more than 92% of Fortune 500 companies are using some iteration of its products. Microsoft, which has to date been OpenAI’s primary strategic investor, sells the AI company’s technology to businesses. But so does OpenAI itself. The company rolled out its specific ChatGPT Enterprise product last year, and has been making progress with partners like Moderna. It said its enterprise team comprises more than 200 people. 

Anthropic has a product team of 100 people in total, about half of which are working on scaled deployment cases such as enterprise, Krieger said.  

But many enterprises are still in their exploratory phase when it comes to AI, something which could benefit Anthropic, he said. 

“I think that gives us an opening to at least be in the consideration set,” Krieger said, adding that every time a new model comes out, it also gives Anthropic a chance to get in front of customers with a trial and see if it solves the problem better. 

Anthropic’s focus on guardrails and explainability—that is, transparency about how a model came to certain outputs—could give it an edge, said Bill Wong, AI Research Fellow at Info-Tech Research Group, an IT research and advisory company. Anthropic was formed by a group of OpenAI employees on that principle. 

According to Krieger, the company’s reputation for responsible AI is typically enough to get the door open, but not enough to fully make the sale. 

When it comes to factors that will differentiate Anthropic’s offering, the Instagram co-founder said he is leaning into the idea of allowing collaboration and social elements while using AI. Up until now, it has typically been one user putting in a prompt and getting a response out. But Anthropic now offers the ability to let teammates have joint conversations with Claude based on shared information about a given project, he said. 

“I definitely saw the value of social features with Instagram, where somebody is like, Oh, I’m learning from other people,” Krieger said. “Obviously, we won’t build exactly that for Anthropic and for Claude, but really leaning into, what does it mean to not feel like alone in your LLM usage, but feeling like you’re collaborating and making things better?”

Write to Isabelle Bousquette at isabelle.bousquette@wsj.com