Something seems to be going on. It so happens that I have recentlly also started to contemplate using Claude more, i.e. to pay for Claude Pro. I have always somehow liked it but rarely used it because it is such a pain to log in (email link) and you get logged out pretty quickly. That’s not the only reason, though.
I have free access to ChatGPT Edu at work (also Microsoft Copilot, but I never use it because it’s crap compared to ChatGPT). I also have a year of free access to Perplexity as well as Gemini AI Pro (if you work in academia: go ahead and get both. It says “students” but works for staff too). I also pay for Kagi Search (which also has AI built in). Well and I also have free access to Github Copilot (again: Edu), which also includes Claude Sonnet. I also love and use NotebookLM a lot.
So, the reasonable mind would say: why would you want to pay for another LLM? Well, by using Github Copilot, I repeatedly noticed that Claude is not only faster but also better than ChatGPT. And it doesn’t praise me for my supersmart questions all the time. Just a few days ago, in the middle of a coding project (hobby), I ran out of premium credits on Github Copilot, so I switched to ChatGPT, but I couldn’t get the integration with VS Code to work properly, so I considered Claude Code, but I knew and that if I used Claude Code, my 5 USD API balance would be used up in no time. So I switched to gemini-cli (free access to Gemini 2.5 Pro). At first it worked fine, but then dragged me into this rabbithole of nonsense that I then tried to make it fix or revert and it just kept making things worse, saying I will now do this but then doing something completely different. I wont go into more details, I suppose we all have had such experiences when the LLM just looses its mind while still sounding competent. But this cost me at least six hours of my time and in the end I had to restore the latest working version from git because it was just such a mess.
I then went to Claude Code and gave it about the same task as Gemini six hours earlier and it just did exacty what I wanted within minutes. And I don’t thing that I was just lucky, because (as others have noted) Claude transparently made a plan of what it was going to do (and that included familiarizing itself with the existing code base, something both Gemini and ChatGPT don’t seem to do unless you explicitly ask them to) and then Claude ticked off one box after the other, provided a perfect summary of what it did, no jargon or hyperbole, I ran the code and it worked.
I gave it some more tasks and it completed all of them flawlessly. It understood the project (even though it was developed in a different conversation), wrote code when I wanted code and answered questions when I asked questions (Gemini, at one point, just kept reproducing the same code when I asked it to explain what a particular section of the code did). That session of 1-2 hours cost around 5 USD, but it was totally worth it. That cost is not sustainable on a daily basis, though, so that is why a Claude Pro subscription became an option.
We’ll see if I actually take the step, given that an hour ago, my monthly allowance of premium requests on Github Copilot has just been refilled, but what you guy’s wrote about using Claude for non-coding tasks reminded me of earlier experiences I’ve had with it and made me want to use it more.
I’m not sure, though if I would say that I’d settle for a specific product or provider. Settling, in my mind, would mean that I can by a yearly subscription. But that is out of the question, these days, with the industry moving so fast.
