And was it worth it? What were you able to achieve with the LLM-support you payed for? How much time did you save?

I did experiments with Claude Sonnet and Opus and also with Gpt-5, the latter via my Perplexity subscription. My experience with Claude was mixed; most output required significant re-work which mostly consumed the time savings enabled by Claude. In contrast, Gpt-5 was able to generate code for several complex problems and made different transpilations for me, and the generated code usually compiles and correctly runs up front, all covered by my Perplexity subscription.