Claude 3.5 Sonnet Drives API Migrations

Claude 3.5 Sonnet turned coding into Anthropic’s wedge into the developer stack. In practice, developers do not buy a model for abstract intelligence, they buy the model that writes the fewest broken edits across large codebases and stays cheap enough to sit inside high frequency products like Cursor, v0, and Replit Agent. That is why model quality translated directly into API share gains, not just brand heat.1 2 3 4 5

• Anthropic paired strong coding performance with product features developers actually feel, especially long context windows and prompt caching. That made it easier to feed whole repos, docs, and previous turns into the model while keeping costs down, which is exactly what AI coding products need to work reliably at scale.1 5
• The migration mattered because the winners in AI coding were heavy API consumers, not occasional chatbot users. Cursor grew from $1M ARR in 2023 to $65M by November 2024, then to $200M by March 2025, while Replit Agent helped push Replit to $70M ARR by April 2025. When those products standardized on Claude, Anthropic captured their usage growth.2 3 4
• This also clarified the split with OpenAI. OpenAI kept the bigger consumer business and far larger ChatGPT distribution, while Anthropic became the model lab most tightly wired into coding workflows and work oriented software. One company owned mindshare at the app layer, the other gained ground where developers were generating tokens all day.1 5

The next phase is vertical integration around coding. Once a lab becomes the default model inside AI IDEs and app builders, it can move up from selling tokens to owning the full coding agent, while competitors respond by buying distribution or launching their own agents. That pushes the market from best model wins to best model plus workflow wins.4 5 6