API Key vs Subscription: Which One Do You Actually Need?
Plain-English guide for a small firm: what an API key is, why your chat plan does not cover automation, and why adding a second person does not double the bill.
A subscription and an API key are two separate accounts with two separate bills. Most small firms only find this out when they try to automate something and the vendor asks for a key they have never heard of.
The rule is short. You need a subscription when a human is sitting there chatting. You need an API key when nobody is watching and software does the work on a schedule.
This guide covers what a key actually is, why the two billing systems never touch, what a key costs to run a small automation, and the safety basics that keep you out of trouble.
Short answer: API key or subscription?
You need a subscription when a person is chatting with the AI, and an API key when software calls the AI on its own. Most firms end up needing both, and that is normal rather than a mistake.
Your monthly chat plan covers one human logging in and typing. It does not extend to a script, a scheduled job, or an automation tool acting for you.
If nobody is at the keyboard when the work happens, the work needs a key.
- A human types the prompt → subscription
- Software sends the prompt on a schedule → API key
- A tool like Zapier or n8n does the sending → API key
- Both at once is the common setup for a small firm
Opening a developer account and wiring an API key for your first automation, and unsure how it should bill next to your firm's chat plan? We will set the account, the key, and the spend cap up correctly in one short session.
Book a ConsultationWhat is an API key, in plain English?
An API key is a long password that identifies your business to the AI vendor's computers. Software uses it to prove the request is yours, the same way your login proves it when you chat.
You create the key once inside the vendor's developer account, then paste it into whatever tool needs to call the AI. Nobody logs in and nobody types a prompt by hand.
The key also decides who pays. Every request carrying that key lands on that account's bill.
- A long secret string, created in a developer account
- Used by software instead of a username and password
- Identifies your business and routes the charges to you
- Created once, then pasted into the tool that needs it
Why the AI vendor has two websites and two bills
Every major AI vendor runs two products on two different sites: a chat app for people and a developer platform for software. OpenAI splits them across chatgpt.com and platform.openai.com. Anthropic splits them across claude.ai and its developer console.
The two sides do not share a wallet. Signing in to the chat app with your work email gives you nothing on the developer platform, and the developer platform never checks whether you have a chat plan.
So a firm paying a flat monthly rate for a business chat plan still has a zero balance on the developer side. Nothing is wrong with the account. It has simply never been opened.
- Chat side — a monthly plan, per person, for humans in a browser
- Developer side — a separate account, billed by usage, for software
- The same login can exist on both sides with separate billing
- A paid chat plan grants no developer credit at all
Why automation cannot run on your chat subscription
Automation cannot use your subscription because a subscription is tied to a person logging in, and automation runs when nobody is logged in. Your 2 a.m. job has no seat to sit in.
That design is deliberate. Chat plans are sized for human pace, so a script firing thousands of requests would blow past what a monthly seat is priced for.
The scheduled routines we run across our own portfolio of sites all bill against an API key, never anyone's chat plan. Nobody is at a keyboard when they fire, which is exactly why the vendor bills them separately.
- A subscription follows a person; automation has no person
- Chat plans are priced for human-speed, hands-on use
- Scheduled jobs run overnight, with nobody signed in
- Every automation tool asks for a key, not your chat password
Does a second person double the API bill?
No. API billing has no seats, so adding a second person to your firm does not add a second charge. You pay for the work sent, not for how many people exist.
The bill is measured in tokens, which are the small chunks of text the model reads and writes. A sibling guide covers how tokens are counted in detail.
So two partners sharing one automation pay the same as one partner running it, as long as the volume of work is the same. Double the volume and the bill doubles. Double the headcount and it does not.
- Subscriptions: priced per user, per month
- API keys: priced per token used, with no user count
- Two people, same workload → same API bill
- Same person, double the workload → double the API bill
Subscription vs API key, side by side
The two options split cleanly by who is doing the asking, and the table below shows where each one wins. Neither replaces the other.
Read the rows in order and your own answer usually appears by the third one.
| Subscription | API key | |
|---|---|---|
| Who it is for | A person who chats, drafts, and asks questions | Software that runs a task for you |
| How it bills | Flat monthly fee | Per token of text sent and returned |
| Do seats matter | Yes — you pay per person | No — there are no seats |
| Nobody is watching | Does not work; needs a login | Works; this is what it is built for |
| Cost when unused | You still pay the monthly fee | You pay nothing |
| Where you get it | The chat app's plan page | The vendor's developer console |
| Verdict | Buy it for every person who chats daily | Buy it once for the firm, for everything automated |
- Subscription wins for daily human drafting and questions
- API key wins for anything scheduled, triggered, or unattended
- Most small firms end up paying for both, on purpose
What does an API key cost to run?
Creating an API key costs nothing. You are charged only for the text your automation sends and receives, and a small automation usually lands in the range of a few dollars a month.
The published rates are per million tokens. Anthropic charges $1 per million input tokens and $5 per million output tokens for Claude Haiku 4.5, its cheapest model. OpenAI charges $0.20 per million input tokens and $1.20 per million output tokens for the lowest GPT-5.6 tier.
Run the arithmetic on a real job. Say an automation reads 200 client emails a month and writes a short summary of each: roughly 400,000 tokens in and 60,000 tokens out. On those Anthropic rates that is about $0.70 for the month, and on those OpenAI rates about $0.15.
- Creating a key: free on every major vendor
- Claude Haiku 4.5: $1 per million input, $5 per million output (Anthropic)
- Cheapest GPT-5.6 tier: $0.20 per million input, $1.20 per million output (OpenAI)
- A small monthly automation: usually cents to a few dollars
- Idle month: no usage means no bill
How do you keep an API key safe?
Treat the key like a password to a payment account, because that is what it is. Anyone holding it can spend your money, and the vendor will bill you for it.
Set a monthly spend cap the day you create the key. Both major developer consoles let you set a hard limit, which turns a runaway script into a stopped script instead of a surprise invoice.
Never paste a key into a chat window, an email, or a shared document. If a key does get out, delete it in the console and create a new one, which takes about a minute.
- Set a monthly spend cap before you use the key
- Store the key in your password manager, not a spreadsheet
- Never paste a key into a chat, email, or shared doc
- Use a separate key per tool, so you can revoke just one
- Delete and replace any key you think has leaked
How to decide for your firm this week
Write down each AI task you do, then mark whether a person is present when it happens. That one column tells you which account to open.
Tasks with a person get subscription seats, one per person who uses AI daily. Tasks with no person get a single firm-level API key, shared by every automation you build.
Then check the numbers before you commit. Our AI model cost calculator turns your expected volume into a monthly estimate, and the model pricing page lists current rates by vendor.
- List every AI task and mark 'person present' yes or no
- Yes → a subscription seat for that person
- No → one firm API key for all of it
- Estimate the volume before you pick a model
- Set the spend cap, then build the first automation
Frequently Asked Questions
- A subscription is a monthly fee that lets a person chat with the AI in an app. An API key is a credential that lets software call the AI directly, billed per token used. Different accounts, different bills, different jobs.
- Because your business plan covers people signing in and chatting, not software running on its own. Automation calls the vendor's developer platform, which is a separate account with separate billing. Your chat plan gives it no credit.
- No. API billing is per token of text processed, with no seats and no per-user fee. Two people sharing one automation pay the same as one person running it, as long as the amount of work is the same.
- The key itself is free. You pay only for usage, at published per-million-token rates — for example, $1 per million input tokens and $5 per million output tokens for Claude Haiku 4.5 (Anthropic). Small firm automations typically run from cents to a few dollars a month.
- No. ChatGPT Plus covers the chat app for one person. Automation goes through OpenAI's separate developer platform, which needs its own account and its own API key, billed by usage.
- Yes, if you handle it like a password. Paste it directly into the tool's credential field, set a monthly spend cap in the vendor console, and use a separate key per tool so you can revoke one without breaking the rest.
- Usually yes. The key only serves software. If anyone at your firm drafts, researches, or asks questions in the chat app every day, that person still needs a subscription seat.
- Your API bill goes to zero for that month. API billing charges for usage only, so an idle key costs nothing, unlike a subscription seat that bills whether you use it or not.
Not sure which account your automation should bill to?
Book a free 30-minute AI workflow audit with Layer3 Labs. We will map which of your tasks need chat seats and which need a single API key, then size the monthly cost before you spend anything.
Book Your Free AI Workflow Audit