Anthropic beats OpenAI as the top LLM provider for business - and it's not even close

oxygen/Getty

ZDNET's key takeaways

• Programming is AI's killer app.
• The top business AI, especially for programming, is Anthropic. 
• Open-source AI is lagging behind its proprietary competitors.

If you were to ask J. Random User on the street what the most popular business AI Large Language Model (LLM) is, I bet you they'd say OpenAI's ChatGPT. As of mid-2025, however, Anthropic is the leading enterprise LLM provider, with 32% of enterprise usage, according to Menlo Ventures, an early-stage venture capital firm.

Before you get too excited, though, keep in mind that Menlo Ventures is a major Anthropic investor. The firm has backed the company through several significant funding rounds, including leading their Series D round and participating in their $3.5 billion Series E, which valued Anthropic at $61.5 billion. 

Also: What happened when Anthropic's Claude AI ran a small shop for a month (spoiler: it got weird)

In other words, Menlo Ventures has billions of reasons to praise Anthropic. That said, others also view Anthropic as the top enterprise AI company. As AI Magazine put it, "Anthropic has established itself as the premier enterprise AI company through its Claude family of LLMs, achieving remarkable 1,000% year-over-year growth to reach $3 billion in annual recurring revenue." Even by hyper-aggressive AI standards, that's real growth.  

Behind Anthropic, you'll find OpenAI, which now has 25%; Google with 20%; and Meta Llama with 9%. All the way in the back, with a mere 1% you'll find DeepSeek, followed by the rest of the pack. Menlo Ventures credits Anthropic's rapid ascent to the strong performance of its Claude Sonnet and Claude Opus models.

Menlo Ventures

(Disclosure: Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

These numbers reflect the proportion of production AI use, not spending. They were derived from a survey of 150 technical decision-makers at enterprises and startups building AI applications in the summer of 2025. 

What's behind Anthropic's success?

Three different factors are driving Anthropic's rise. The first is what Menlo Ventures calls "AI's first killer app": Code generation. While AI-created code quality remains questionable, nevertheless, more developers are using AI programming tools than ever, and Claude has become programmers' top choice with 42% of the market share. That's double OpenAI's 21% share. 

There are concrete examples of Anthropic development programs gaining popularity. For instance, in just one year, Claude helped transform GitHub Copilot into a $1.9-billion ecosystem. Claude Sonnet 3.5's 2024 release showed how LLM breakthroughs can make possible entirely new categories such as AI IDEs, Cursor and Windsurf; vibe app builders, Lovable, Bolt, and Replit; and enterprise coding agents, Claude Code and All Hands. 

Another reason Anthropic is winning is its use of reinforcement learning with verifiable rewards (RLVR) to train its LLMs. Behind that complicated name lies a simple concept: You provide clear, binary feedback (1 for correct, 0 for incorrect) on the model's output. This works well for programming AI tools, where the code either works or doesn't.

Anthropic has also led the way to LLMs that take step-by-step approaches to solving problems and use external tools to pull in data to deliver better answers. In short, Anthropic has been a leader in creating AI agents. Besides helping people and programmers, this approach can help LLMs iteratively improve their responses and integrate tools like search, calculators, coding environments, and other resources via the Model Context Protocol (MCP). This new open-source protocol enables LLMs and AI agents to seamlessly connect with the vast, ever-changing landscape of real-world data, tools, and services.

Also: 7 strategic insights business and IT leaders need for AI transformation in 2025

That's important because Menlo Ventures also found that it's not price that drives companies to change LLMs, it's performance. "This creates an unexpected market dynamic: Even as individual models drop 10x in price, builders don't capture savings by using older models; they just move en masse to the best-performing one."

This dynamic may change once LLMs start to mature and models begin to reach similar performance levels. For now, though, as LLMs improve massively from one release to another, companies are willing to pay for the newest and fastest. 

AI in the enterprise

The study also found that companies are steadily shifting from building and training models to inference, that is, with models actually running in production. Startups are leading the way, with 74% of builders now stating that most of their workloads are in production. Large enterprises aren't far behind, with 49% reporting that most or nearly all of their computers are in production. In short, enterprises are now using AI, not merely building AI.

Finally, the researchers said that open-source LLMs have declined to 13% of AI workloads today from 19% six months ago. The market leader remains Llama, albeit that Llama isn't really open source.

Also: How agentic AI is transforming the very foundations of business strategy

Nevertheless, more open-source LLMs have been appearing. These include new models from DeepSeek (V3, R1), Bytedance Seed (Doubao), Minimax (Text 1), Alibaba (Qwen 3), Moonshot AI (Kimi K2), and Z AI (GLM 4.5) in the last six months. 

They're just not used much. That's because, despite their advantages, "greater customization, potential cost savings, and the ability to deploy within private cloud or on-premises environments," their performance has continued to "trail frontier, closed-source models." Add in that many of the best-performing open-source LLMs to date are from Chinese companies that Western businesses are wary of, and open-source LLMs appear to be stalling out.

Also: Anthropic tripled its revenue in 5 months - and this is why

So what does all this lead to?

"Predicting the future of AI can be a fool's errand. The market changes by the week, with exciting new model launches, advancements in foundation model capabilities, and plunging costs," Menlo Ventures said. Still, "conditions are ripe for a new generation of enduring AI businesses to be built on top of today's foundational building blocks."

The question remains, however, "What will those foundational building blocks be?" OpenAI? Google, Meta? Anthropic? Stay tuned. We're not yet close to being able to say which AI models will ultimately end up on top.   

Get the morning's top stories in your inbox each day with our Tech Today newsletter.