I Pay for 5 AI Subscriptions Every Month—Here’s Why That’s Actually Not Stupid

Last Wednesday, a mate pinged me on WhatsApp with a question that stopped me mid-scroll.

“Which AI do you actually use the most these days?”

I stared at my phone. Opened my home screen to grab a screenshot. Five icons—ChatGPT, Claude, Gemini, DeepSeek, Grok—sitting in a folder I’d labelled “AI Tools” like it was the most normal thing in the world. I actually laughed out loud.

I don’t use one the most.

I use all of them.

Here’s the thing—and this took me over a year of burning cash to figure out—asking “which AI is best” in 2026 is like asking whether a saloon, an SUV, or a pickup truck is best. Best for what? Moving house? The school run? A track day?

They’re not even the same category of thing. They just happen to all speak English.

Let me break this down. No hype. No fanboy nonsense. Just what I’ve learned from actually using these tools every single day.

ChatGPT: The Reliable One (That’s Been Getting on My Nerves)

I’ve been using GPT the longest—since late 2022. Four generations later, my overall impression is one word: solid.

Writing, coding, translation, research—no obvious weak spots. Image generation and video are both top-tier now. Not the absolute best at either, but comfortably in the top two. The 5.5 model’s coding ability? Decent. Not mind-blowing, but decent.

But.

The Plus plan has been getting noticeably dumber lately. They’ve slashed the codex quota on 5.5 dramatically. Mid-conversation on the web version, it’ll silently downgrade your model—you won’t even know it’s happened. Upload four or five images and suddenly you’ve hit some invisible limit.

Seriously, CloseAI? Are things that tight?

And the web interface. God. It stutters and lags like it’s running on a potato. The $200/month Pro plan unlocks the full-fat o3 model, and honestly? I pay it. But every renewal stings. I wince. Every. Single. Time.

Look, if you’ve only got budget for one subscription, ChatGPT is still your safest bet. It’s the most well-rounded, no question. But here’s its real problem: it doesn’t sound human.

GPT’s responses are thorough, precise, and dripping with textbook energy. Smooth. Agreeable. Painfully balanced. So I’ve developed a habit—actually, it’s muscle memory at this point—where I let it draft, then toss the output to Claude or Gemini for a human-sounding polish.

This workflow is baked into my DNA now.

Claude: God-Tier for Code and Writing (But Increasingly Picky)

Claude used to be my go-to for anything code-related.

During the Opus 4.6 era, its programming ability was miles ahead. Left other models in the dust. Then 4.7 started… dribbling. And 4.8? 4.8 is really leaning into it.

What do I mean by “dribbling”?

I mean the model’s started making stupid mistakes. Basic ones. Its information retrieval is especially weak—far behind GPT—in scenarios that require precise understanding. It’ll take a simple request and overcomplicate it to death, burning 20 minutes to produce something that’s somehow further from what you asked for than when it started.

Honestly. The harder it tries, the more it misses the point.

But—and this is a big but—Claude is still untouchable for creative writing and deep discussions. Its prose has texture. Emotional range. It doesn’t sound like it’s wearing a suit and tie. Its Chinese language processing has improved massively too, now trading blows with DeepSeek depending on the task.

It’s expensive, though. The Max subscription isn’t cheap, and the free Sonnet tier burns through its quota in about two minutes. You blink, it’s gone.

I’ve paid. You haven’t. That’s why you’re stuck in traffic—harsh, I know, but that’s roughly the dynamic here. If you don’t write code, the free 4.6 tier is probably enough.

Gemini: Ecosystem Magic (Just Don’t Ask It to Do Maths)

Gemini’s the one I’ve been reaching for more and more over the last six months.

Why? Because it’s deeply woven into Google’s ecosystem, and my workflow lives in Google Docs, Gmail, and YouTube. In those contexts, Gemini’s convenience is unmatched. Nothing else comes close. Its multimodal capabilities are also top-tier—95.8% image understanding accuracy, video analysis that can handle up to three hours of continuous frame processing. Uploading images barely counts against any limit. For day-to-day chatting, it’s genuinely pleasant.

Brilliant.

But—and this “but” is doing some heavy lifting—the 3.5 Flash model has been lobotomised into a complete idiot. And 3.1 Flash-Lite? Lobotomised into a slightly smaller idiot.

Giving either of these serious work?

Pure, unfiltered chaos generator.

It can’t even do basic arithmetic reliably. And there’s something else that really winds me up: Google uses your conversations for training data. Don’t want that? Fine. Then you don’t get chat history. Severely crippled. It’s basically the American version of a budget AI with no voice functionality.

So here’s how I use it now: heavy lifting goes to 3.1 Pro. Casual chat uses Flash-Lite (it’s free, whatever). Anything that matters? Flash doesn’t touch it.

This allocation strategy took three separate disasters to figure out.

DeepSeek: China’s Dark Horse (But Let’s Not Get Carried Away)

DeepSeek has been the biggest surprise of 2025-2026.

No debate there.

With the V4 release, it’s undergone a genuine transformation. It’s basically a mini-GPT now. In maths, coding, and logical reasoning, it matches or even edges past GPT-5 in some benchmarks. For Chinese language comprehension and composition? It’s in a league of its own. Untouchable.

And it’s free. The API pricing is absurdly low—roughly 1/10th of GPT’s rates. Ten quid’s worth of credit lasts ages. Proper bargain.

But don’t get swept up in the hype.

Early versions were rambly, unreliable, and basically useless for serious work. V4 is much better, but it still lacks multimodal capabilities—no image generation. Complex tasks still make it stumble.

My take: if you don’t have easy access to Western services and just want one solid model, DeepSeek is more than enough. Hook it up through Cherry Studio via API for better performance and fewer restrictions. For text-based needs, it’s cheap and perfectly capable for daily use.

And here’s something to chew on: all those resellers in China have been hyping up specific models since the 3.5 Sonnet days—first for coding, then writing, now Gemini—just to get you to pay for their relay services. The gap between domestic and international models exists, but it’s tiny now. You really don’t need to pay 20-30x the price for a marginal difference.

Think about it.

Grok: Regret in Subscription Form

Oh, Grok. I just sigh now.

Ten dollars a month for SuperGrok Lite. The only genuinely useful feature? Its search—it taps directly into X (Twitter) data, so hunting down real-time information and trending topics actually works well. It’s also slightly more… relaxed about certain content boundaries. Not that it’s a free-for-all.

Beyond that? Nothing.

Information retrieval is weaker than GPT. Writing is painfully stiff. Hallucinations are rampant. It struggles to understand basic requests. I think I got swept up in the Elon mystique—subscribed on impulse.

Mistake.

Definitely cancelling next month. If you’re hyper-sensitive to real-time info, use the free tier for casual searches. The API isn’t worth your time. Don’t bother.

Here’s the Actual Strategy: Use Them Together

My current daily setup looks like this:

• Workhorse: ChatGPT (general tasks, translation drafts)
• Polisher: Claude (creative writing, code, making things sound human)
• Google ecosystem: Gemini 3.1 Pro (docs, video, casual chat)
• Chinese & value: DeepSeek (domestic tasks, API calls)
• Real-time search: Grok free tier (occasional use)

They’re not competitors. They’re complementary.

This is the bit most AI tutorials won’t tell you—because they’re trying to sell you a course on one specific tool.

What I still can’t figure out: why does every company nerf their free tier? GPT gets dumbed down, Claude gets quota-squeezed, Gemini gets lobotomised, and Grok was never smart to begin with.

Feels coordinated.

Whatever. They’re businesses. I get it. Still annoys me.

TL;DR / Key Takeaways

• ChatGPT: Most balanced. Best all-rounder if you can only afford one. But it sounds like a textbook.
• Claude: Unmatched for creative writing and nuanced expression. Coding ability has declined but still strong. Expensive.
• Gemini: Ecosystem integration is unbeatable. Pro model is solid. Flash models are useless for anything serious.
• DeepSeek: Incredible value. Chinese language performance is best-in-class. No multimodal yet, but for text work it’s brilliant.
• Grok: Good for real-time X/Twitter search. Everything else? Don’t pay for it.

It’s 2026. AI tools are everywhere, but most people still don’t know how to actually use them effectively. If you’re stuck wondering which one to subscribe to, try my approach: figure out what you spend 80% of your time doing, then subscribe to the model that’s strongest in that specific scenario. Everything else? Free tiers will do.

Don’t be like me. Don’t subscribe to all five.

Hope this saves you some cash. I’ve already made enough expensive mistakes for the both of us.

What’s your AI stack looking like these days? Found any combination that works surprisingly well? Drop a comment—I’m genuinely curious.

AI #ChatGPT #Claude #Gemini #DeepSeek #Productivity #TechReview