OpenAI Platform Overview

Last updated: 2026-08-12

Overview

OpenAI Platform is the control plane for API development: issue keys, view bills, try models in Playground, and jump to official documentation. Many integration failures aren’t “wrong code” but time lost on wrong blogs, old endpoints, or expired model IDs. This guide maps console tasks ↔ doc sections ↔ code fields so you find answers by task. UI changes—verify against what you see after login.

Platform vs ChatGPT web: don’t mix them

Dimension	chatgpt.com	platform.openai.com
User	End user	Developer
Billing	ChatGPT subscription	API by token / add-ons
Credential	Account login	API key
Output	Chat experience	Programmable HTTP calls

Integrations use Platform only. Try ChatGPT web for conversation feel; key creation, usage monitoring, and production calls still depend on Platform.

Console features by development task

Task	Where	Practical tip
Create / revoke keys	API keys	Separate keys for dev / staging / prod
Spend, alerts	Usage / Billing	Estimate per-request cost with real prompts before launch
Try models and prompts	Playground	Confirm in Playground before moving to code
Available models	Models (docs or console)	Example IDs in tutorials may be outdated
Team permissions	Organization / Projects	Isolate when multiple products share one account
Troubleshooting	Logs (if available) + x-request-id	Distinguish 401 (key) from 429 (rate limit)

If Platform is temporarily unreachable, read Dev Overview for concepts; keys and billing still require Platform access.

How to read the docs site in 30 minutes

Suggested order for platform.openai.com/docs:

1. Get started / Quickstart — current primary endpoint (often Responses API)
2. Models — context length, capabilities, price tiers
3. Guides — text generation, tool use, RAG, structured output
4. API reference — authoritative field definitions; wins over Stack Overflow
5. Changelog / Deprecations — Assistants and other sunset timelines

If Assistants is marked legacy: compare Assistants guide with Responses API.

API keys: creation through rotation

# .env for local dev; never commit to Git
export OPENAI_API_KEY="sk-..."

# Quick key check (model list means auth succeeded)
curl https://api.openai.com/v1/models \
  -H "Authorization: Bearer $OPENAI_API_KEY"

Practice	Why
Keys per environment	Limits blast radius on leak
Key naming / labels	Know who uses what during rotation
Dual-key gray rotation	Zero-downtime switch
Never expose in frontend	Browser = public

Key usage and production checklist: API Developer Guide.

Billing, rate limits, and cost estimation

• Token billing: input and output priced separately; system, RAG chunks, and tool schemas count as input.
• 429 rate limit: RPM/TPM exceeded; use exponential backoff, not unlimited concurrent retries.
• Pricing: openai.com/api/pricing — record usage from Playground or test scripts, multiply by unit price.
• Batch API (if available): offline jobs may be cheaper—see Batch in docs.

Parameter and token optimization: Text Generation guide.

Playground → production workflow

Playground works → copy request JSON → move to server
       ↓                                      ↓
  pick model / system                   add timeout, retry, logging
       ↓                                      ↓
  record token usage                  golden set compares output

Note: with temperature > 0, Playground and API outputs won’t match byte-for-byte; evaluate structure and coverage, not string diff.

Organization, projects, and data policy

• Organization ID: some SDK flows and support tickets need it; large teams allocate quota by project.
• Data usage policy: check official API data usage—training use, retention duration.
• Regional restrictions: if account or IP hits region limits, read support docs; don’t violate ToS to bypass.

Related resources

Resource	When to use
Cookbook	Copyable RAG, agent recipes
openai-python	Official Python SDK
status.openai.com	Check before blaming your code on widespread 5xx
API pricing	Selection and budget

Frequently asked questions

Are Platform and ChatGPT the same account?

Usually the same OpenAI account system, but API billing and ChatGPT subscription are separate.

Does Playground cost money?

Generally consumes API quota; confirm in Billing.

Can’t find Assistants in docs?

It may have moved or been marked legacy. Read Assistants API guide and Deprecations.

Does 401 always mean a bad key?

Often missing key, revoked key, or wrong env var name—also check you didn’t use a ChatGPT session token.

Model works in Playground but 400 in code?

Request body fields vary by API shape; verify model and messages/input structure in API reference.

Official resources

• OpenAI Platform
• API documentation
• API pricing
• ChatGPT

Next reading

• OpenAI Dev Overview
• OpenAI API Quickstart
• ChatGPT API Developer Guide
• Prompt Engineering

Action path

Today: Log into Platform, create a local-dev key, write it to .env. Tomorrow: Find the Quickstart primary endpoint; run once in Playground and once with curl. This week: Set a billing alert and complete server-side calling from Quickstart.