OpenAI heavily invested in the code tool Cursor, but it has now switched sides by quietly changing its default model to Claude.
When registering a new account and installing Cursor without modifying any settings, the first code completion is handled by Claude-3.5-Sonnet.
Anysphere, the company behind Cursor, received $8 million from OpenAI in its early stages, accounting for 72.7% of that funding round. This made them appear to be firmly in OpenAI's camp.
Their switch to a different model indicates that OpenAI has temporarily lost dominance in the important code generation scenario. It may also reflect that OpenAI's investment agreements are relatively flexible.
Cursor has become very popular recently, with many programmers canceling their GitHub Copilot subscriptions in favor of Cursor, despite Cursor being twice as expensive at $20/month.
Some developers have even gone to great lengths to overcome their VS Code habits, like renaming the Cursor app to VS Code to trick themselves into using it.
Cursor's appeal comes largely from its ability to use Claude-3.5-Sonnet. It also has innovative features like:
- "Cursor position prediction" which automatically predicts where you want to edit next
- Multi-line editing for efficient formatting changes
- The experimental Composer feature for editing across multiple files simultaneously
In benchmark tests, Claude-3.5-Sonnet has been outperforming GPT-4 and ChatGPT in code generation tasks.
Other platforms like Replit are also promoting Claude-3.5-Sonnet integration as a key feature.
Overall, Claude appears to be establishing dominance in the code generation domain, both in model performance and adoption by major development tools.