"OpenAI" isn't one product — it's the company behind ChatGPT, the developer API at platform.openai.com, the video tool Sora, and the coding agent Codex, and each one has its own sign-in flow and its own bill. Once you know which door you actually want, getting in takes about two minutes.
Short answer: Go to chatgpt.com to chat for free, or platform.openai.com to build with the API — one OpenAI account logs into both. ChatGPT is a flat monthly subscription; the API is pay-per-token and needs a card on file, with GPT-5.6 Sol currently priced at $5 per 1M input tokens and $30 per 1M output tokens. No free API credit is granted to new accounts anymore.
In my testing I set up a fresh OpenAI account from scratch and walked both paths — chatting at chatgpt.com and generating an API key at platform.openai.com — specifically to see where a first-timer gets confused. The confusion is almost always the same: people think a ChatGPT Plus subscription includes API access, and it doesn't. Here's how the two actually fit together, how to get a working API key, and the mistakes that waste the most time.
What you'll need
An email address, or a Google, Microsoft, or Apple account, is enough to create an OpenAI account — no card required just to sign up. That single login works for both chatgpt.com and the developer dashboard at platform.openai.com, so you don't create two separate identities. If you only want to chat, stop there; nothing else is needed. If you want to call the API from your own code, you'll additionally need a payment method on file before OpenAI raises your rate limits past a bare trickle of requests, according to OpenAI's own developer quickstart. A modern browser covers everything in this guide — no software install required to test either surface.
Step-by-step: how to use OpenAI
1. Decide which OpenAI you actually want
If the goal is "ask it questions, get writing help, generate images," that's ChatGPT at chatgpt.com. If the goal is "build a feature into my own app or script," that's the API at platform.openai.com. They run on the same underlying models but are billed and accessed completely separately — this is the single most common mix-up I see.
2. Create one OpenAI account for both
Go to chatgpt.com or platform.openai.com and sign up with an email or an existing Google, Microsoft, or Apple account. Whichever you start from, the same login works on the other site — you don't need a second account to move from chatting to building.
3. Start chatting at chatgpt.com if that's all you need
Sign in, type a question into the message box, and press enter — there's nothing else to configure. My full how to use ChatGPT guide covers plan pricing, prompt examples, and the mistakes worth avoiding on that side specifically.
4. Open the API dashboard and create a key if you're building something
Go to platform.openai.com, sign in, and open Dashboard → API keys. Click Create new secret key, give it a name you'll recognize later, and copy it immediately — OpenAI shows the full key exactly once, per its own developer quickstart. Paste it somewhere safe, like an environment variable, not into a chat message or a shared doc.
5. Add a payment method before you build anything real
OpenAI's quickstart documentation describes a single free test request to confirm your key works, then explicitly points you toward adding credits "to keep building." In my testing, a brand-new key without billing attached could confirm the connection worked but hit a wall almost immediately on anything beyond that. Add a card under Settings → Billing before you plan a real project around it.
6. Test a prompt in the Playground before you write any code
Inside the dashboard, open the Playground to send test prompts, compare models, and check token costs without writing a line of code. It's the fastest way to confirm a prompt works the way you expect before wiring it into an app.
Example prompts you can copy
Whether you're testing in the Playground or chatting at chatgpt.com, a prompt that states its format and constraints up front beats a vague one every time. A few I reused across both surfaces while testing this guide:
- "Summarize this document in 5 bullet points, then flag any number or date I should double-check: [paste text]."
- "Act as a code reviewer. Point out the biggest risk in this function before suggesting any rewrite: [paste code]."
- "I'm choosing between [option A] and [option B] for [use case]. Give me the three real tradeoffs, not a generic list."
- "Return the answer as JSON with keys
summary,risk_level, andnext_step— nothing else in the response." - "Rewrite this for a reader who knows [related topic] but has never seen this specific idea before."
That last two matter more on the API side than in casual chat — being explicit about output format is most of what separates a reliable API prompt from one that breaks your parser on the tenth call.
Common mistakes to avoid
The costliest one I ran into is assuming a ChatGPT Plus subscription pays for API usage — it doesn't; they're separate products with separate bills, and a Plus subscriber with no card on the API dashboard still can't make real calls. Second, losing an API key because it's only shown once at creation — if you didn't copy it, the fix is generating a new one, not hunting for the old one in your dashboard. Third, skipping the Playground and debugging a broken prompt directly in production code, which makes every fix slower than it needs to be. Fourth, not setting a usage limit or billing alert under Settings → Billing, so a bug that loops API calls turns into a surprise charge instead of a capped one. And fifth, treating every OpenAI product as the same thing — Sora, Codex, and the core API each have their own access rules and pricing, so what's included in your ChatGPT plan doesn't automatically carry over.
OpenAI's tools at a glance
| Tool | Where | Billing | What it's for |
|---|---|---|---|
| ChatGPT | chatgpt.com | Free plan, paid plans up to $200/mo | Chat, writing, research, images |
| API / Platform | platform.openai.com | Pay-per-token; GPT-5.6 Sol is $5/$30 per 1M input/output tokens | Building your own apps and scripts |
| Codex | Inside ChatGPT and the CLI | Included with ChatGPT Plus and up | Coding agent, multi-file changes |
| Sora | sora.com | Included with some ChatGPT plans or separate credits | Text-to-video generation |
GPT-5.6 Sol's $5-input/$30-output per-million-token rate, plus a $0.50 cached-input rate, comes straight from OpenAI's own API pricing page — worth a direct check before you budget a project, since per-token pricing shifts more often than a flat subscription does.
Tools that make this easier
If chatting is genuinely all you need, my full how to use ChatGPT guide breaks down what each paid plan actually adds, and how to use ChatGPT for free covers exactly how far the no-cost tier stretches before you'd notice a limit. For multi-step, tool-using tasks rather than a single question, how to use ChatGPT agents covers what Agent mode can complete on its own, and how to use ChatGPT Codex is the one to read before you point OpenAI's coding agent at a real repository — though it's not the only option worth testing, as my ChatGPT alternatives for coding roundup shows. If you're deciding between OpenAI and a competitor before you commit any billing details, I compared the assistants directly in how to use Claude AI and how to use Gemini — in my experience the three are close enough on plain chat that the real decision usually comes down to which ecosystem, Google's, Anthropic's, or OpenAI's, you're already living in. For long-form writing at volume specifically, a dedicated tool can still outperform any general chat assistant; my best AI writing tools guide covers the ones worth paying for.
My take
Using ChatGPT is genuinely a two-minute task — sign in, type, get an answer. The part that actually trips people up is everything OpenAI built around it: an API with its own account state and billing, a coding agent, a video generator, all sharing one login but none of them sharing a bill. Start with whichever door matches your actual task, keep the API key somewhere safe the moment you generate it, and set a billing limit before your first real API project rather than after a surprise invoice. That order kept my own test account from any nasty surprises.
Frequently Asked Questions
Is OpenAI free to use?
ChatGPT has a real free tier — chat, limited image generation, and more at no cost, no card required. The API is different: you can send one free test request, but sustained use needs a payment method on file, per OpenAI's own developer quickstart.
How long does it take to start using OpenAI?
A couple of minutes for ChatGPT — sign up, type a message, get an answer. Getting a working API key takes about the same time, though adding a payment method and testing your first real call usually adds another five to ten minutes.
What is the easiest way to use OpenAI?
If you just have a question, chatgpt.com is the easiest path — no setup beyond signing in. If you're building something, test your prompt in the Playground first rather than wiring it straight into code, so you're debugging the prompt and the integration separately.
Do I need a credit card to use the OpenAI API?
Not to sign up or send an initial free test request, but yes for anything beyond that. In my testing, a key with no billing attached hit its limits almost immediately, which matches OpenAI's own guidance to add credits before building further.
Is a ChatGPT Plus subscription the same account as the API?
Same login, separate billing. A Plus subscription unlocks more inside chatgpt.com; it does not add any API credit or usage to platform.openai.com. You pay for API calls separately, per token, regardless of which ChatGPT plan you're on.