Three months ago I opened my credit card statement and noticed I'd spent more on AI tools than on groceries. That's not an exaggeration. It was a Tuesday in January and I just sat there looking at the number.
$90 in AI subscriptions. In one month.
I'd accumulated them gradually enough that none of it felt dramatic at the time. $20 here for ChatGPT Plus. $20 there for Claude Pro. Cursor was $20 more. GitHub Copilot, which I kept telling myself I was "evaluating," was another $10. And then there were the smaller ones — a $12 tool I used twice, a $9 thing that I couldn't even remember signing up for.
So I started tracking.
What I was actually paying
When I put everything into a spreadsheet in January, this is roughly what I had:
- ChatGPT Plus — $20/mo
- Claude Pro — $20/mo
- Cursor Pro — $20/mo
- GitHub Copilot — $10/mo
- Perplexity Pro — $20/mo (forgot I had this one)
- A few smaller tools — another $20-25 scattered around
Total: somewhere around $110-115 per month.
Annualized, that's close to $1,400. For someone running side projects and not billing those tools to a client, that's a lot.
The uncomfortable part: I wasn't using most of them daily. ChatGPT and Claude were doing maybe 80% of my actual work. Cursor was genuinely useful but probably not $20-useful when VS Code + a good free extension could cover a lot of the same ground. Copilot I was barely touching because Cursor had taken over for me.
Perplexity I had literally forgotten about.
The audit process
I went through each subscription and asked a simple question: what specifically did I do with this tool in the last 30 days that I couldn't have done with something else?
For ChatGPT Plus, the answer was clear — o1 for debugging tricky problems, image generation occasionally, and it's just my default. Keeping it.
Claude Pro was harder to justify as a separate $20. I was using it mostly for long-context document work and writing. But I wasn't using it every single day. More like a few times a week.
Cursor Pro I decided to keep because the autocomplete saves me real time and the diff-based editing is something I actually depend on now. But I looked at whether I was using it at the scale that justified the pro tier. Probably, yeah.
Copilot had to go. The overlap with Cursor was too high, and I wasn't getting $10 of incremental value out of it. Cancelled.
Perplexity I also cancelled. The free tier is genuinely good enough for the way I use it.
What actually helped: rethinking the structure
The bigger unlock wasn't cancelling things. It was figuring out how to pay less for the things I wanted to keep.
Annual billing on Claude went from $20/month to roughly $18 when you spread the annual plan across 12 months. Not huge, but it adds up over a year. ChatGPT has a similar structure.
Student discounts are real and most people don't think to check. If you're in school or have a .edu email, ChatGPT Plus, Notion, Perplexity and a bunch of others have steep discounts — sometimes 50% off. GitHub Copilot is free for verified students through the Student Developer Pack. That's not nothing.
Shared subscriptions are a thing that not everyone talks about openly but plenty of people do. Tools like ChatGPT Plus, Claude Pro, and Perplexity Pro have family or team-style plans where costs split across multiple people. There are services that specialize in connecting people who want to share subscription costs on AI tools. Worth knowing about, though I'll be honest — the shared model does come with tradeoffs. You don't always get the same priority access during peak hours, and sometimes feature rollouts are slower on shared tiers. I tried a shared Perplexity arrangement for a month and found the latency occasionally annoying.
Free tiers that are genuinely good enough. This one surprised me. Claude's free tier is more capable than it was a year ago. If I was just starting out and wanted to explore AI coding assistance, I'd start with free Copilot via GitHub Student Pack, free Claude for general chat, and free Perplexity for search. That stack costs nothing.
My current stack and what I actually pay
After three months of tracking and trimming, here's where I landed:
| Tool | Plan | Monthly Cost |
|---|---|---|
| ChatGPT Plus | Annual billing | ~$20 |
| Claude Pro | Annual billing | ~$18 |
| Cursor Pro | Monthly | $20 |
| GitHub Copilot | Cancelled (overlap) | $0 |
| Perplexity | Free tier | $0 |
| Everything else | Gone | $0 |
Total: around $58/month, down from $110+.
That's saving maybe $600 a year, which is not insignificant. The actual productivity difference from cutting those subscriptions? Minimal. The tools I kept are the ones doing real work.
The thing that helped most was just writing it down. I'd been running on vibes — "I should probably keep this" — without ever forcing myself to articulate the value. Turns out, "probably useful" is not a good reason to spend $120/year on something.
A note on the free tier trap
Free tiers have gotten genuinely better, but they're not free in the sense of unlimited. Claude's free tier will rate-limit you hard if you're doing serious work. ChatGPT's free tier doesn't include o1 or o3, which are the models I actually care about for debugging. Cursor's free tier gives you a limited number of "fast" requests before throttling you.
If you're using these tools for hobby projects or learning, free is probably fine. If AI tools are in your daily workflow and you're actually depending on response quality and speed, the paid tier on one or two tools is likely worth it. Just not five paid tiers.
Pick the one or two that you'd notice if they disappeared tomorrow. Pay for those. Question everything else.
Anyway, that's what the spreadsheet told me. Not revelatory, but it was useful to actually look.
Top comments (0)