Last spring I printed every AI subscription receipt for a single month and taped them to the wall above my desk. Fourteen charges. Eleven tools. £247, most of it auto-renewing without so much as a reminder email. That wall is why this pricing review exists.
Most “best AI tools” lists quote the headline number and move on. Twenty dollars here, fifteen there. That headline is almost never what you pay. Credits expire, seat minimums appear at checkout, annual discounts vanish when you add a second user, and the feature you actually need sits one tier above the one you signed up for. I covered the feature side of this in my roundup of the AI tools I’d genuinely pay for, but pricing deserved a separate piece.
Read the billing unit before you read the price
Every AI product prices something slightly different, and that difference matters far more than the number attached to it.
- Seats: you pay per person, whether or not they log in that month
- Credits: you pay per action, and the vendor decides what counts as an action
- Usage bands: a flat fee until you cross a threshold nobody advertised clearly
- Annual lock-in: the price you saw, conditional on signing away twelve months
A $20 seat and a $20 credit bundle look identical on a pricing page. They behave nothing alike by month three.
Where the flat $20 subscriptions earned their keep
The big general-purpose assistants stayed boringly predictable, which is a compliment. ChatGPT Plus, Claude Pro, Perplexity Pro and Gemini Advanced all sat at roughly $20/month for the year I tracked them. No credit meters, no surprise throttling on ordinary use. I ran a proper cost log during six months of paid subscriptions across a dozen tools, and the flat-fee assistants were the only category where the actual spend matched the quoted spend to the penny.
Two caveats. Annual billing saves around 17% on most of them, but only commit if you’ve already paid monthly for two months and know you’ll keep going. And shared team plans often cost more per person than individual seats, not less, because the team tier bundles admin controls and higher usage ceilings you may never touch.
Watch the model-switching surcharge
Several assistants now gate their strongest reasoning models behind a higher tier. On one tool, moving from the $20 plan to the tier that unlocked the newer model took the bill from $20 to $200 a month. I tested it for a fortnight and cancelled. The 10x price did not produce 10x output on the work I was doing, which was mostly editing and research summarising.
Credit pricing is where the real surprises live
Image, video and voice generators love credits because credits are flexible for them and opaque for you. Three examples from my spreadsheet:
Image generation. The $10 basic tier on one popular tool gave me roughly 200 fast generations. Re-rolls, upscales and variations each burned credit. A design-heavy week ate the month in four days.
Voice. Character-based audio pricing sounds generous until you realise a single 20-minute narration with retakes can consume an hour of generated audio. My $22 plan covered about six finished podcast segments.
Video. This is the most expensive credit market of the lot. Re-rendering a 40-second clip four times at higher resolution cost more than my entire text-tool budget for the month.
None of these are rip-offs. They’re just poorly suited to unpredictable creative work. If your usage swings week to week, favour tools that sell unlimited or high-cap flat plans, even at a higher sticker price.
The cheap tiers that are quietly enough
Some of the most useful subscriptions in my stack cost less than a sandwich, and I keep them precisely because the low tier does the job.
Meeting transcription is the clearest case. The entry plan on Otter covers enough hours for a normal calendar of calls, and I’ve written up how to configure Otter so recordings become follow-up emails rather than a growing pile of unread summaries. The value isn’t in the tier, it’s in whether you wire the output into something you already do.
Video editing follows the same logic. A browser-based editor with a modest monthly cap handled every short clip I needed this year, and the learning curve was close to zero. The expensive suites made sense only for people shipping long-form video weekly.
Per-seat billing and the small-team tax
The most misleading pricing pages belong to automation and agent platforms, because the useful work often happens in the gap between the cheap seat and the expensive one.
Lindy is a good example. The free tier will run a couple of simple automations, and the first paid tier unlocks task volume that a solo operator might genuinely need. Push past that and you’re paying for concurrent runs you probably won’t saturate. I broke the numbers down properly in what Lindy AI costs and where it falls short, and the short version is that per-task pricing punishes messy workflows more than per-seat pricing does.
Developer tooling is the opposite. JetBrains AI Assistant bundles into an IDE subscription you may already hold, so the marginal cost is often zero. My notes on where JetBrains AI Assistant earns its place are less about price and more about whether the included quota matches how much you actually lean on it during a normal sprint.
Three questions that kill a bad purchase
Ask what happens at the cap. Ask whether unused credits roll over. Ask whether downgrading mid-cycle refunds anything. The answers are usually on a help-centre page the pricing page doesn’t link to, and they decide whether a $30 plan is a bargain or a trap.
A ten-minute audit before you renew anything
Open your card statement and match every AI charge to a real week of use. Then run these checks:
- Did I open this tool more than four times last month?
- Am I paying for a tier whose headline feature I never touched?
- Would the free or entry tier cover 80% of my actual usage?
- Is there a duplicate? Two transcription tools, two writing assistants, two image generators
- Does the annual price beat monthly by enough to justify the commitment?
Cancelling one redundant $20 subscription funds the upgrade on the tool you use daily. That trade is almost always worth making.
What a realistic monthly stack costs now
For a freelancer doing writing, research and occasional video, my current setup runs about $70 a month: one $20 general assistant, one $20 transcription plan, roughly $18 for image generation on a mid tier, and a $12 browser-based video editor. A small agency with three people should budget closer to $260 once seat minimums and one automation platform are included.
The pattern worth remembering is that price tracks predictability, not capability. Tools that charge a flat fee are betting you’ll use them moderately; tools that charge per action are betting you won’t. Know which kind of user you are before you click upgrade, and the invoice stops being a monthly ambush.