Which AI model or subscription do you actually pay for, and does it earn its keep

[NightHarbour30]

Plenty of people stick with the free tier of whatever's around, but if you're paying monthly for Claude, ChatGPT, Gemini or something else, what tipped you into actually reaching for your card. Curious whether it's genuinely saving you time or if it's more of a curiosity subscription at this point.

[Sophie92]

There is something satisfying about having a paid AI subscription that you actually use. For me, the value comes from using it for work rather than random questions. Coding, drafting documents, summarising long notes and turning rough ideas into something usable are the big ones.

It still needs supervision, especially with code, but getting from a blank page to a decent first draft is a huge time saver. For casual questions, though, the free tiers are usually plenty

[Bayley_US]

Claude is the one I pay for at the moment. Long-form writing is where it earns its keep for me, particularly when I already have a pile of notes and need help turning them into something coherent.

The useful bit is being able to provide a lot of context without spending half the conversation explaining what every document means. That matters more to me than benchmark scores.

Would I pay for it if I only used AI once or twice a week? Probably not. At that point the free options are good enough.

[Always_Myles26]

Gemini has survived my various AI subscription clear-outs. A lot of that is because I already use Google's ecosystem, so having an assistant that fits naturally around documents and everyday tasks is convenient.

The model quality matters, but workflow matters more to me. If I can move from an email to a document to a summary without constantly copying things between different apps, that is a genuine benefit.

I still switch models for difficult questions, though. We are definitely in the era of keeping two AI tabs open and asking both when something matters.

[Scarlet Annie]

No paid subscription here. That is not an anti-AI stance; I just do not use one consistently enough to justify another monthly bill.

For coding, brainstorming and the occasional explanation, free services cover most of what I need. When I hit a usage limit, I either wait or try another model. Mild inconvenience, but not exactly a disaster.

I can absolutely see the value for somebody using AI every workday. If it saves several hours a month, the subscription can pay for itself pretty quickly.

[Merchant31]

There is a slightly annoying assumption that the most capable model automatically gives the best value. I do not think that follows.

For my use, a fast and reasonably good model is often preferable to a slower premium one. If I need five headline ideas or a summary of some meeting notes, I do not need the AI equivalent of a supercomputer contemplating the universe.

The premium model earns its keep when the task is difficult enough that the extra reasoning or context actually changes the result.

[EventHorizon25]

My employer pays for my work AI account, so technically I pay nothing, but it has changed how I work enough that I would probably subscribe personally if that disappeared.

The boring tasks are where it shines. Turning bullet points into a first draft, cleaning up documentation, extracting action items from meetings, writing test cases and converting one format into another all remove little bits of friction.

The important part is remembering that the output is a draft. I have caught enough subtle mistakes that I would never paste generated material into something important without checking it.

[Beta]

I cancelled my premium AI subscription after about four months. Not because it was bad, but because I realised I was using it to avoid making tiny decisions myself.

Every email became a prompt. Every short message needed three rewrites. I would ask it to compare restaurants, headphones, notebooks and approximately seventeen varieties of pasta sauce. At some point I had outsourced the exciting intellectual challenge of choosing lunch.

Now I use the free tier for genuinely useful stuff and keep a bit more of the mundane thinking for myself. Weirdly, I do not miss the subscription as much as I expected.

[Stu96]

I pay for ChatGPT and use Claude's free tier as a second opinion. That combination works better for me than trying to find one model that wins at absolutely everything.

ChatGPT gets most of my everyday use because I have built up habits around it. Claude gets called in when I want a different take on a long piece of writing or when the first answer feels suspiciously neat.

Sometimes the two disagree, which is actually useful. If two systems confidently give different answers, that is a giant flashing sign saying that I should check an actual source before relying on either one.

[Poppy51]

For me the subscription pays for itself through one very specific use case: studying technical material.

I can take a dense explanation, ask for a simpler version, then ask for a quiz, then deliberately give a wrong answer and have it explain where my reasoning went off track. That interactive loop is much more useful than reading another static tutorial.

It is not a replacement for textbooks or proper sources, especially when precision matters. But as a patient tutor that will explain the same concept five different ways without getting annoyed, it is remarkably handy.

[Cipher31]

The killer feature for me is not raw intelligence. It is continuity.

Having an assistant that already understands the conventions of a project, the terminology I use and the way I like documents structured means I spend less time writing setup prompts. The difference is subtle in one conversation but huge after several weeks.

That is also why I am more cautious about switching providers than benchmark charts might suggest. Starting fresh with a theoretically better model can be less useful than staying with the one that already fits my workflow.

[Dank]

I subscribe to more than one, which sounds ridiculous when I type it out. In practice, though, I use them for different things, so it has stopped feeling like paying twice for the same service.

One is my general assistant for research, coding and everyday questions. The other is mainly for writing and large documents. I also keep a few free models around for quick disposable tasks where I do not care which system answers.

The trick is remembering to cancel things when your usage changes. AI subscriptions have somehow become the new gym memberships: very easy to sign up for, strangely easy to forget about

[FinnHalliday]

Still on the free tier, and I suspect that is partly because I enjoy making things harder for myself.

For normal questions I have not found the paid versions transformative enough to make me reach for my card. If a model can explain something complicated, help me understand a document or suggest dinner from whatever is in the fridge, I am happy.

Where I get tempted is image generation and other advanced tools. Once you use those regularly, free limits can become pretty restrictive. That is probably more likely to make me subscribe than a slightly smarter text model.

[HiggsField]

The funny thing is that my answer changes every few months because the models keep improving. A task that needed a paid tier last year can sometimes be handled perfectly well for free now.

That makes me reluctant to become a brand loyalist. I would rather judge the current product on what it actually does for me than defend something because I picked it six months ago.

For now I pay for a general-purpose subscription, but if the free tiers keep getting better, I could easily see myself dropping it. The monthly bill still has to justify itself.

[Sigma]

The subscription earns its keep almost entirely through programming. Without that use case, I would probably cancel it.

A typical session starts with something painfully mundane: a bug report, a stack trace and a codebase I have not looked at in six months. Having the model explain what it thinks is happening, suggest debugging paths and generate a small test case gets me moving much faster.

There is a catch, though. AI can make bad code look extremely respectable. I have accepted a beautifully written solution that completely misunderstood one requirement. The confidence is free; the debugging bill is not.