I've spent over $3,000 on ChatGPT, Claude, and Gemini, and only one was worth the money

Something I remember particularly well about the early days of generative AI is how quickly people developed an aversion to the idea of paying for yet another subscription. I could understand why! We were already paying monthly for everything from streaming services to cloud storage, and suddenly, using a chatbot just to draft a quick email or (unfortunately) write an essay was going to cost another $20 a month.

In fact, when ChatGPT Plus first launched in February 2023, Reddit was filled with people questioning whether a chatbot could possibly be worth that much. Fast forward a few years, and I've somehow managed to spend over $3,000 on subscriptions to ChatGPT, Claude, and Gemini alone.

After spending that much money, I've come to realize that paying for an AI subscription isn't necessarily a waste of money. Paying for the wrong one, however, absolutely can be! So, here's the only one I'd happily continue paying for...

ChatGPT Plus is the easiest subscription to justify

OpenAI has finally won me over

If you'd told me six months ago that I'd be recommending a $20 ChatGPT Plus subscription over a $100 Claude Max subscription, I'd have looked you right in the eyes and laughed. However, over the last few weeks, ChatGPT has undeniably jumped to the very top of my list, and it's not even a particularly close competition anymore.

For starters, OpenAI's models have gotten significantly better. I've noticed the difference most when using Codex and Claude Code side by side, especially when giving both tools the exact same task. There was a time when Claude Code would consistently deliver better results, particularly when it came to building polished interfaces and figuring out what I wanted without needing me to spell out every little detail. Lately, though, the tables have turned. Codex has been doing a much better job of understanding my intentions, making sensible decisions on its own, and delivering finished products that actually feel finished.

Even when I preferred Claude for coding, I was always vocal about how ChatGPT still excelled as a general-purpose AI assistant. Whether I needed help researching a topic, understanding a complicated concept, brainstorming ideas, or simply figuring something out, ChatGPT was almost always the first tool I'd turn to. Now that Codex has caught up with, and in some cases surpassed, Claude Code in my own testing, I'm finding fewer and fewer reasons to pay for a separate Claude subscription.

The biggest reason, though, is the limits. Claude frankly has absurd usage limits, and I manage to hit my weekly limit just three days into the week! On the contrary, I can get through an entire week of using Codex on my $20 ChatGPT Plus subscription without running into nearly as many restrictions. Considering I was paying five times as much for Claude Max, it's becoming increasingly difficult to justify spending that kind of money when the cheaper alternative lets me get more work done.

Claude has lost the edge it once had

Oh, how the tables have turned

While I've already addressed most of the reasons I don't prefer Claude nearly as much as I used to, I think the biggest one is that it no longer has the unique edge that once made it so easy to recommend. It's not necessarily that Claude has gotten significantly worse. Rather, competing LLMs have caught up with, and in some cases surpassed, the things Claude used to do better than everyone else. And when that happens, all the little frustrations you've been willing to overlook suddenly become a lot harder to ignore.

I also feel like Anthropic has sort of slowed down in the last few months. I've always praised the company for staying ahead of the curve and introducing features that genuinely change how people use AI, rather than simply chasing whatever its competitors are doing. Lately, though, it feels like OpenAI has been moving at a much faster pace, rolling out new features and improvements that give me more reasons to keep coming back. That's not to say Claude has stopped improving, but for the first time in a long while, I find myself getting more excited about what OpenAI is doing than what Anthropic has up its sleeve.

Unfortunately, it's not just the pace of innovation that's been bothering me. A couple of months ago, I wrote an entire article about how exhausting Claude had become to use, particularly for research. Despite using its most capable model at the highest effort setting, I kept finding that Claude would rush through research tasks, overlook useful sources, or answer questions without even bothering to search the web.

In one experiment, I gave Claude and ChatGPT the exact same research question. Claude spent roughly a minute and a half putting together its answer, while ChatGPT spent around 12 minutes digging into the topic. The difference in the quality and depth of their responses was enough to make me question what I was paying so much more for.

Then there's the fact that Claude has occasionally taken it upon itself to decide how I should use my remaining allowance. I've had Claude Code discourage me from continuing a task because I was approaching my usage limit, even though I was perfectly happy to use whatever I had left. I understand warning users before they run out, but I don't think a tool I'm paying $100 a month for should be making that decision for me.

None of this means Claude is a bad AI assistant. Far from it! There are still things I prefer about Claude, and I wouldn't hesitate to recommend trying it. But recommending Claude and recommending a paid Claude subscription are two very different things. With the competition getting better and the frustrations piling up, I just don't see enough reasons to keep paying for Claude Max anymore.

Gemini is the weakest reason to pay for Google AI Pro

Everything but the chatbot makes it worthwhile

Google AI Pro is a slightly different story. Unlike ChatGPT Plus and Claude Pro, where the main selling point is access to better AI models and higher usage limits, Google's subscription comes bundled with a bunch of other perks that make it surprisingly easy to justify.

For instance, I use NotebookLM practically every day for studying, and I genuinely wouldn't hesitate to pay $20 a month for its premium features alone. The higher limits are incredibly useful, especially when I'm working with multiple lectures, readings, and presentations. So, the fact that Google AI Pro includes an upgraded NotebookLM experience alongside 5TB of cloud storage, Gemini integrations across Google's productivity apps, and access to other tools like Flow and Antigravity makes it a pretty compelling package.

There's even YouTube Premium Lite included in select countries, which is a nice bonus if you're someone who spends a ridiculous amount of time watching YouTube. And if you already rely heavily on Gmail, Google Docs, and Drive, I can absolutely see Google AI Pro being the subscription that makes the most financial sense for you.

The problem, at least for me, is Gemini itself. While Google's models have certainly improved, I still find myself turning to ChatGPT or Claude for most of the things I'd normally use a general-purpose AI assistant for. Whether it's brainstorming ideas, researching something, or tackling a complicated task, Gemini rarely ends up being my first choice.

In every comparison where I've pitted Codex, Claude Code, and Antigravity against each other, Antigravity has, without fail, been the one that disappointed me the most. It's not necessarily that it can't get the job done, but the results often feel far less polished than what its competitors deliver. I've had instances where Antigravity declared a project finished despite leaving out features I'd explicitly requested, and others where I had to step in to fix problems that ideally should've been caught during testing.

That's ultimately why I find Google AI Pro a little difficult to recommend purely for Gemini. The subscription itself offers tremendous value, particularly if you're already invested in Google's ecosystem, but I don't think the chatbot or coding experience alone is compelling enough to make me choose it over ChatGPT Plus.

So, all that to say, the subscription I can most easily recommend right now is ChatGPT Plus. Claude would be my next pick if you're looking for an AI assistant primarily for writing or coding, though I'd think twice before splurging on its more expensive tiers. I'd only recommend Google AI Pro if the AI itself isn't your main reason for subscribing, and you're more interested in perks like NotebookLM, cloud storage, and integrations with Google's other services.