I can’t even remember how many times I changed my choice of chatbots in everyday use. Every time one chatbot angered me, I wanted to replace it, but each chatbot has its own limitations and weak points, so in the end I’d remain with nothing in use! Based on ever-changing criteria, I tried to justify to myself the choice of the moment, only to be proven wrong not long after.

If I were to make a comparison, chatbots are as inconstant in performance today as antivirus solutions were in the past: today, this one and that one are great; six months later, one is a sieve, and the other is a memory and CPU hog. One year later, everything changes.

With chatbots, or rather with the underlying LLMs, the situation is even worse. The more their makers declare each new version as better than the previous one (benchmarks, benchmarks, benchmarks!), the more people complain that the respective models are dumber and dumber.

For now, I revised my Android apps and my browser bookmarks to a new set of chatbots, and I’ll try to explain (if that still matters) my decision.

I didn’t forget my previous arguments, but I also added new ones, sometimes contradictory:

  • The Chinese chatbots deserve to be boycotted.
  • I need to have several chatbots “ready to use” precisely because of the increasing usage limits for both free and paying users!
  • Gemini helped me a lot recently, but now the new limits are quite severe, and many people decided to boycott it. (Of course, there are ways of getting an 18-month Pro subscription for $15 + €1.99. It’s not illegal, but there’s the risk of an abrupt termination.)
  • People complained and keep complaining that “Gemini got dumber,” “Claude got dumber, and it blocks accounts believing they’re used by kids” (this is random but still happens), “ChatGPT got dumber.”
  • A recent search for an unknown wristwatch (only by a description) highlighted Grok, Qwen, and Kimi as the only helpful ones, albeit not entirely so.
  • Grok is now severely limited for free users: only Fast.
  • ChatGPT is, in my experience, almost unusable. It’s dumb, there is no way to know what model it’s using, and despite being configurable in the answering style, I still cannot stand it.
  • Copilot, with its many, many flaws, it still lets you choose between Smart, Think deeper, Study and learn, and Search. It’s based on GPT, but it’s a better choice for free riders than ChatGPT, once you accept that there’s no way to get fewer icons and bullets.
  • Mistral is increasingly concise and dumb, at least if you don’t subscribe. I hoped it would improve, but I see no sign of this happening.

So, well, err… voilà:

  • Gemini is not that bad. It seems to have decreased in quality, but it’s still very useful for a wide range of topics. (It can also generate images.)
  • Grok is also good to have, even with decreased depth of thinking (if there’s still any of it). (It can also generate images with the least censorship of all.)
  • Copilot is sometimes incredibly annoying or plain stupid, but it still has its uses. I kept testing it, so I know when it could be of use. (For images, too, but it tends to censor.)
  • I can’t keep Kimi and Qwen aside once I know they have their fortes. I already have a long history with Kimi, and occasionally Qwen gives quite smart answers. Good to have.

I don’t need to love, trust, or recommend any of them. That’s fine, because I don’t. What I need is to have some tools at hand.