The Pulse: ‘Tokenmaxxing’ as a weird new trend

The Pulse: Interesting AI coding stats from Cursor

Cursor recently released a report based on two years of data from its AI coding platform, highlighting some key trends. Top 1% of users generate 30-40K lines of code per week, equivalent to 45 median users. The report also found that input tokens (used for reading codebase and documentation) comprise 90% of Cursor's token usage, with output tokens taking up only 0.6% when caching is considered. The efficient usage of input tokens over output tokens suggests that context reuse and caching will be a crucial aspect of future AI efficiency. The report also compared the cost of different AI models, finding that Opus 4.7 is about 10x more expensive than Composer 2.5, but produces similar results.