"Subscriptions are very expensive, there is no room to experiment" — what trying it actually costs
"The cost of AI is high, subscriptions are very expensive, there is no room to experiment" — a verbatim quote from the survey, and the person is right. To try three models you need three subscriptions, each charged in full. Below: why that breaks the entrance and what a different payment model changes.
"The cost of AI is high, subscriptions are very expensive, there is no room to experiment" — a verbatim quote from the survey, and the person is right. To try three models you need three subscriptions, each charged in full. Below: why that breaks the entrance and what a different payment model changes.
Three subscriptions you barely use
The objection from my survey, verbatim: "the high cost of AI, subscriptions are very expensive, there is no room to experiment".
The person is right on the main point: under a subscription model experimenting genuinely is unaffordable. To try one model for video, a second for images and a third for text, three subscriptions have to be taken out. Each is charged in full, every month, regardless of whether you opened it.
What follows is predictable. You take out one — the best known — and try to do everything in it. It goes badly, because each model has its own strengths. And the conclusion the person draws is about himself: "it is not working, I suppose I have not figured it out".
But it is not working for another reason. You are using one tool where three are needed — because there is no money for three.
And the second half of the problem: it is unclear what a specific task will cost. You do not know in advance what a clip will come to — you only know the subscription was charged.
What is wrong with subscription as a model
The root cause: a subscription is payment for access rather than for work. You pay for the right to use it, and pay the same whether you made two images that month or two hundred. For somebody with a flow of work that is good value. For somebody trying it out it is a pure loss.
The misconception: "I will buy the subscription first and start earning after". The order is the reverse. A subscription pays off when there is a flow. While there is no flow, it takes money and adds guilt: paid for and not using it.
And a systemic layer. For most services a subscription is more profitable than per-use payment precisely because some people pay and do not use it. That is not a conspiracy, it is a business model — but it is worth knowing about.
How to count differently
There is another model, and it removes the objection entirely: a wallet instead of a subscription.
Payment on usage. Money is debited for work done rather than for a calendar month. Did not open it for a week, spent nothing.
The estimate is visible before you press. You see the cost of an operation before launching rather than after. That changes behaviour itself: you can compare two options by price and choose, rather than guess.
Topping up in small amounts. The entry threshold is a few dollars rather than a monthly subscription. With that money you can try five different models instead of one.
The difference is like a gym membership against taxis. A membership is good value for somebody who goes three times a week. For somebody who goes once a month it is cheaper to pay per journey.
What is available with no money at all
Separately, because people usually do not know this.
The catalogues are open to guests. Descriptions of skills, prompts, assistants and templates are visible without registering — you can see what exists before spending anything.
The blog is fully open, no registration needed.
Free registration gives three days of full Basic access. Not a demo and not a cut-down version but the full plan: chats with every assistant, the prompt texts, the skill files, six foundational express courses and the ten-lesson sales module.
Three days is enough to work through “Freelancer packaging” and come out with a finished package of documents: a niche, a client avatar, a portfolio and profile texts. That is, to test not the tool but yourself — before paying for anything.
Experience converts into access. Advanced skills and templates open either with a plan, or one-off for $9, or for 1 000 experience points earned by working on the platform.
What one piece of work actually pays
Worth looking at the spending from the other side — from the side of what the work pays.
By the Parselona feed — 5 986 jobs over 4 months — the median one-off job is around $60. A short clip $18–45, a logo $25–90, a multi-page site $245–730.
Now set that against the cost of generation. With per-use payment the cost of an operation is measured in cents and single dollars rather than tens. That is, the very first job completed covers the cost of several dozen attempts.
The whole problem is the order. A subscription demands money before the first job. Per-use payment lets you start with a sum you do not mind and then pay out of what you earn.
How to spend sensibly when there is money
Three rules that save more than any choice of plan.
One tool until the first payment. Until you have earned, new services are not tried at all. Every new tool is not only money but time, which at five hours a week you do not have.
Count the spend per task, not per month. The right question is not "what does the subscription cost" but "what did this job cost me". If the job brought $40 and the spend was $3, the model works. If it is the other way round, the price of the subscription is not the issue.
Do not pay for what you need once a quarter. A one-off payment for the operation you need is almost always cheaper than a year of access for two uses.
Three more ways not to overpay
Free tiers of the models cover more than it seems. For text tasks, drafts and testing an idea, paid access is often not needed at all. Paying makes sense where volume begins, or quality that goes to a client.
Do not pay for something already included in something else. People regularly take out a separate subscription to a tool they already have access to inside something larger. Before paying, check what you do not already have.
Count the spend as part of the job's quote. When you name a price, the cost of generations goes into it the way paper goes into a printer's. A contractor who pays for tools out of his own pocket and does not build it into the price is working at a loss and does not notice.
What to do today to give it a try
Check your current spending: open your list of subscriptions and look at how many times you opened each over the past month. Usually it turns out the money goes on two or three and the work happens in one.
Then start with the free option rather than with choosing a plan. Registration costs nothing, and three days of the Basic plan switch on with one button, no card needed. Basic includes over 200 AI assistants (marketing, sales, legal, HR, e-commerce, analytics, creative, engineers), the "Sales" course, six Basic mini-courses including “Freelancer packaging”, “Freelance 2.0” and an encyclopaedia of 198 verified AI services.
The task of those three days is not to learn everything. The task is to choose one service, one tool and one platform and get to a first application. Everything else is bought later, out of money earned rather than instead of it.
Having no money for subscriptions is a real constraint. But it constrains experimenting rather than earning: to take a first job you need one working tool, not five.
If you know somebody paying for three AI services and earning nothing on any of them, forward this to them.
The order of prices by operation
So that "pay on usage" does not sound abstract — here is what the operations themselves cost.
| Mode | Models | Cost per generation |
|---|---|---|
| Images | 30 | $0.12 – 0.25 |
| Text to video | 51 | $0.22 – 11.95 |
| Photo → video | 84 | $0.17 – 11.95 |
| Photo editor | 34 | $0.12 – 0.50 |
| Upscale | 6 | $0.12 – 0.40 |
| Background removal | 1 | $0.12 |
| Product Shot | 2 | $0.12 |
| Lip Sync | 13 | $0.12 – 1.50 |
| Video editor | 39 | $0.12 – 11.95 |
| File → text | 4 | $0.15 |
What is visible immediately. Static costs pennies: twenty images is a few dollars. Video is the only category with a fiftyfold spread, and that is exactly where quoting mistakes get made.
What a finished clip costs in total
A reference point for a completed piece of work rather than for one generation.
A clip is assembled from several generations: frames, animation, voiceover, assembly. Hand that to Cerebrum (on the Full plan the AI SKILLS connector for Claude does the same): you describe the task in plain words, the estimate for all the steps is visible before launch, and the total is simply the sum of the prices in the table above.
Compare with the market: a short clip sells for $18–45. The difference is your work.
How payment works on the platform
Everything that generates runs from the wallet: top up from $5, debited on usage, the price visible before you press the button.
Gen AI — 338 models in one place instead of a dozen subscriptions. Images from $0.12, video from $0.22, file transcription $0.15. No standing charge for idleness: no generating, no paying.
The AI SKILLS connector for Claude (Full plan) — the same generations straight from a Claude chat. The estimate is shown before launch, and money for anything not delivered returns to the balance automatically: if a step did not complete, its cost comes back.
Cerebrum — the chat on the platform: the model's estimate, the cost of the move and the balance on one line before launching. Nothing paid without explicit agreement.
The balance does not expire — topped-up money stays on it until you spend it on generations.
What needs no wallet at all
Assistants — over 200 roles across many directions, Basic plan. Open a chat, give a brief, get a result. No installation, no payment per generation.
Skills — files with Basic, advanced ones with Creator+, or $9 one-off, or 1000 XP. Experience accrues from learning: a lesson 10 points, a course 100, an express course 50, a webinar 20. So advanced content opens with no money at all.
Registration is free, and three days of the Basic plan switch on with one button, no card needed.
How to experiment without paying for subscriptions
The article is about a wallet being cheaper than a subscription. Here is the second half of the answer: part of the work need not be paid for at all. All three skills below are advanced: Creator and up, or $9 one-off, or 1000 XP.
- quant.cpp — compressing a model's memory more than sixfold: a 128-thousand-word context fits on a laptop with 16 gigabytes
- Token and cost audit — where the money goes, broken down by model. It also shows that a cheap model saves over ninety percent on some tasks
- Splitrail — the same in real time, across several tools at once
Where the real saving is. Not in the choice of plan but in splitting the tasks. Drafts, iterations and experiments go to a local or cheap model; the expensive one is kept for the final result. The audit from the list above shows the difference in figures — and it is large enough to be worth changing the habit for.
What to pick up on the platform for this task
- Claude Code cost tracking — a skill: what you spend on Claude Code by day and session; advanced (Creator+ / $9 / 1000 XP)
- A gateway for your own LLM keys with auto-optimisation — open source: your own keys with no markup, the gateway lowers costs by itself; description open to all, repository link from Creator
- Jan — offline AI chat on your own computer — open source: a model with no subscription on your own hardware; repository link from Creator
- Strategic Consultant — how to reach a first job on AI skills; from the Basic plan
- AI tools encyclopaedia — 198 verified services in 24 categories so you do not buy a redundant subscription; $9, Basic plan and up
What to read next
LLM pricing and token economics — how the cost of a request is calculated.
The cost of producing a clip — what a real job comes to.
Which plan to choose — an analysis by task rather than by tick boxes.
It was expensive, now it is $9 — what changed in the access model.
What you can do from a phone — where the boundary runs if there is no computer.