Anthropic's API Closing on OpenAI
Claude Code vs. Cursor
This marked a real shift in where model demand was being created, because Anthropic was no longer just selling a strong model, it was becoming the engine underneath fast growing coding products that generated heavy usage every day. By September 2024, Anthropic had reached about $664M API ARR, up 5x year to date, while OpenAI’s API business was at about $1.08B, up 2x, so the gap was still large but closing quickly.
-
Anthropic won on developer friendly features that mattered in practice. Bigger context windows let apps pass in more code and docs at once, and prompt caching cut repeat query costs sharply, which made Claude easier and cheaper to use inside products like coding copilots and app builders.
-
The revenue mix shows the difference clearly. By March 2025, about 80% of Anthropic’s $1.4B ARR came from API usage, versus OpenAI where 73% of revenue came from ChatGPT and 27% from API. Anthropic was more concentrated on developers and downstream software products, while OpenAI was more consumer led.
-
That is why Cursor, Bolt.new, v0, and Replit mattered so much. These products turned model quality into recurring token spend. Every accepted edit, generated file, and debugging loop pushed more usage through Anthropic’s API, creating a compounding distribution channel that looked different from OpenAI’s chat first growth engine.
The next phase was always going to be labs moving up the stack to capture more of that workflow value. Anthropic’s launch of Claude Code showed that winning the API layer in coding could become a bridge to owning the end product too, which pushed the market toward tighter competition between model providers and the applications built on top of them.