That's interesting, b/c that research from a few months ago seems to line up with the article. The article actually calls out 38x when you compare CC with Cursor's API token pool only. But then when you add in the Composer tokens, it becomes 5x. Obviously Composer's a different level of model than Sonnet/Opus.
"In a confession... that is sure to raise new questions about its privacy policies..." The data wasn't private! Questions should not be posed about a nonexistent point.