I Cancelled ChatGPT Plus, Claude Pro and Google AI Pro. Here's What I Use Instead.

2026-09-12 · 8 min read

chatgpt-plusclaude-progeminialternativespricing

For about fourteen months I paid for all three: ChatGPT Plus at $20, Claude Pro at $20, and Google AI Pro at $19.99. Sixty dollars a month. Seven hundred and twenty a year, which is more than I spend on every other piece of software I use, combined.

I didn't sign up for three subscriptions on purpose. Nobody does. It happens one justification at a time.

This is what the fourteen months actually looked like, why I cancelled, what I replaced them with, and — the part most of these articles skip — the two things I genuinely lost in the process.

How I ended up with three

ChatGPT Plus came first, in the way it does for most people. It was the default. I used it for everything for about six months and it was fine at everything.

Claude Pro came second, and for a specific reason: long documents. I was working through contracts and research papers, and Claude held the thread across 40 pages in a way GPT at the time didn't. I told myself it was temporary. It was not temporary.

Google AI Pro came third, and this is the one I'm least proud of. It arrived attached to storage I already wanted, bundled in a way that made the marginal cost feel like $6 instead of $20. Gemini was genuinely strong on anything that touched my own Gmail and Docs. I was also, if I'm honest, curious.

At no point did I sit down and decide to spend $720 a year on chat interfaces. Each individual charge was defensible. The total never got evaluated, because the total never appeared anywhere — three separate cards, three separate renewal dates, three separate receipts filed in three separate months.

That's the actual mechanism. Subscription spending doesn't grow by bad decisions, it grows by unexamined ones.

The month I actually looked

What forced the reckoning was mundane: I was categorising expenses for the year and three AI line items landed in the same spreadsheet column for the first time.

So I did the thing I'd been avoiding and went back through my history across all three, month by month, and tried to work out what each subscription was actually earning.

Here's roughly what I found:

ChatGPT Plus. Heavy use, but overwhelmingly on tasks that didn't need a flagship model — rewording emails, quick factual lookups, formatting messy text. Real value was concentrated in maybe two things: image generation, which I used a few times a month, and voice, which I used more than I expected while walking.

Claude Pro. Lower volume, higher value per use. The long-document work was real and Claude was genuinely better at it. But it was maybe six to eight sessions a month. I was paying roughly $2.50 per session for the privilege.

Google AI Pro. Almost entirely the Workspace integration. The chat interface itself I barely opened — when I wanted to talk to a model I went to ChatGPT or Claude out of habit. I was paying $20/month for what amounted to a very good sidebar in Gmail.

And then the thing that didn't show up in any single account, because it was spread across all three: I was doing a lot of manual multi-model work. Ask Claude, paste into GPT, ask whether it agrees. Get a draft from GPT, run it past Claude for a second read. That cross-checking was probably the single most valuable thing I did with AI all year — and it was also the most tedious, because it meant copy-pasting between three tabs that knew nothing about each other.

I was paying $60/month for three walled gardens and then doing the integration work by hand.

What I actually cancelled, and in what order

I didn't cancel all three at once. I'd recommend not doing that — the point isn't to save money, it's to find out what you were really buying.

Google AI Pro went first, and I've never thought about it again. The Workspace integration was the only thing I used, and I dropped to the free tier, which still covers the Gmail and Docs work at a lower ceiling I rarely hit. Saved: $19.99/month, felt: nothing.

Claude Pro went second, and this one I planned. Claude Fable 5 is available through several multi-model services, and I moved my long-document workflow to one of them (AIWITH.CHAT — I'll be specific about it below since that's the honest thing to do) and ran a month in parallel before cancelling. Same model, same outputs, different window.

ChatGPT Plus went last, and took the longest, because that's where the habits lived. It wasn't a capability problem. It was fourteen months of muscle memory pointed at one URL.

What I use now

One subscription: AIWITH.CHAT at $9.9/month, which gives me GPT-6 Astra, Claude Fable 5, Gemini 3.1 Pro, DeepSeek V4.1 and Kimi in a single chat window, with a shared credit pool across all of them.

The bill went from $59.99 to $9.9. That's $600 a year, and the saving isn't the interesting part.

The interesting part is what happened to the cross-checking. It stopped being a chore. In one window I can ask Claude Fable 5 for an analysis and then switch models mid-conversation and ask GPT-6 whether it agrees — with the full context carried over, no copy-paste, no re-explaining. The workflow I'd been doing manually for a year became a dropdown.

The second change was subtler and took a few weeks to notice. I stopped routing everything through flagship models. When you pay $20 for one lab, every task goes to that lab's best model because that's what you're paying for. When five models share one credit pool, running a bulk text cleanup through DeepSeek instead of a frontier model is just... the obvious choice. It's cheaper in credits and equally good at the job. I use frontier models for frontier problems now, which is what I should have been doing all along and wasn't, because the billing structure punished it.

The two things I genuinely miss

This is the part I'd want to read, so here it is honestly.

1. Native image generation and voice. ChatGPT's image generation was good and I used it. Voice mode while walking was a habit I liked. Neither transferred. If either of those is core to how you work, cancelling ChatGPT Plus is a real downgrade and no amount of model access compensates for it — be clear-eyed about that before you cancel.

2. Day-one access to new features. When OpenAI or Anthropic ships something new, it lands in their own app first. Aggregators follow, sometimes by weeks. I'm no longer in the first cohort for anything, and if you're someone who needs to be — if you write about this stuff, or your work depends on being early — that lag is a real cost.

There's a third case that doesn't apply to me but applies to a lot of people reading this: if you use Claude Code, keep Claude Pro. It's bundled into the subscription, it's not available through any aggregator including this one, and if you code with it daily it's the best-value $20 in your stack. The maths in this article simply doesn't apply to you.

Would I do it again?

Yes, but I'd do it faster.

The fourteen months of paying triple weren't a mistake of judgement — each subscription made sense when I bought it. The mistake was never re-examining the stack as a whole. Three defensible decisions added up to an indefensible total, and the only reason it persisted is that nothing ever forced me to look at the sum.

If you're paying for two or three of these right now, you don't have to take my conclusion. But do the exercise: open each one, scroll back through a month of history, and ask what each subscription is actually earning. Most people find the same thing I did — one subscription doing real work, one doing habitual work a cheaper model would handle fine, and one they forgot they had.

If you're considering the same move

A few things I'd tell myself fourteen months ago:

Cancel in order of least attachment, not largest saving. You want the easy one first to prove to yourself nothing breaks.
Run a month in parallel before cancelling the one you actually rely on. New accounts get free credits, so this costs nothing but a month of overlap on the subscription you were already paying for.
Check the two exceptions first: Claude Code, and native image/voice. If either is load-bearing for you, stop here — the honest answer is keep paying.
Watch for the habit, not the capability. The last subscription is always the hardest to drop and it's almost never because of what the product does.