anonyfox

Hey guys just a simple question to all of you who heavily leverage LLM coding daily. I essentially have cursor run in 1-3 projects in parallel all day, asking for the next steps implementation like every ~10 +- 5 minutes, and with Claude opus 4.5 it just works for me naturally without much discussions these days, it just flows. But I see like more than 1000€ per week burned in token costs due to that. Currently employer pays for it and it’s mostly work projects, yet I kind of am concerned now that I would really be crippled if I wouldn’t have it on my own if I quit. Programmer for decades so I CAN still write by hand efficiently but it’s just night and day compared to when vibecoding after 2 years became natural and shipping speed like that is the new felt normal.

I tried different models to maybe get pricing savings yet ultimately only Opus 4.5 kinda oneshots my thoughts still. GPT is kind of autistic and useless at times (but good at research and abstract modeling), other models are „nice“ but still make mistakes or need much more guidance that slows down again or become too sloppy with increasing complexity. And unfortunately I seem to be unable to get Grok 4.1 thinking enabled in Cursor, it’s only grok code fast which is amazing for quick edits but complex issues the full grok model is the closest to Claude yet not properly in cursor.

I also haven’t seen the open models be comparable just yet, let alone one that runs fast enough on a M1 air.

So what do you guys do or think here? Everyone just enjoying the benefits of employers handing out access? Or just swallowing the costs? Or not yet there to fully max out usage day to day? It’s a increasingly serious issue in my head, essentially cursor costs per month would be higher than rent in my area, so this affects job retention/considerations already..