API vs Subscription
Subscriptions are for using AI tools. APIs are for building AI features into products.
Subscription
Pay a fixed monthly fee to use AI tools directly.
- ChatGPT / Claude / Gemini apps
- Codex / Claude Code / Cursor
- GitHub Copilot
API
Pay per usage to build AI into your own products.
- OpenAI API
- Anthropic API
- Google Gemini API
- DeepSeek API
Which do you need?
| Use case | Subscription | API |
|---|---|---|
| Personal coding | Good | Sometimes |
| Building a chatbot | No | Yes |
| Internal automation | Sometimes | Yes |
| SaaS feature | No | Yes |
| Cost per user control | Limited | Yes |
If you are using ChatGPT, Codex, Claude Code, Cursor, or Copilot directly, you may not need API pricing.
If you are building AI features into your own product, API pricing matters.
API-key usage is generally billed separately from AI app subscriptions. Eligible third-party apps using Sign in with ChatGPT can instead draw from shared ChatGPT plan usage limits. Check how your tool signs in before estimating a separate API bill. OpenAI billing · ChatGPT in other apps.
Eligible Claude Max, Ultra, and Team plans offer monthly Claude Platform API credits through a linked Console organization. Claude Agent SDK and claude -p use those credits when you run them yourself with an API key from that organization; when signed in with a Claude plan instead, they draw from subscription limits. Credits do not cover interactive Claude Code or increase plan usage limits. Anthropic credit rules · Agent SDK sign-in rules. This calculator estimates metered API costs before credits, not subscription usage allowances.
App-builder examples
Use these as starting shapes before opening the calculator. The token and request volumes mirror the calculator examples guide, and should be edited for your own product.
| Product pattern | Subscription is enough when | API pricing matters when | Start with |
|---|---|---|---|
| Support chatbot | Works for drafting replies manually. | Matters when customer messages trigger your product flow. | 1k input / 500 output x 10k requests Chatbot guide |
| RAG answer feature | Works for testing prompts with pasted context. | Matters when retrieved context is assembled per request. | 4k input / 800 output x 5k requests RAG guide |
| Document summarizer | Works for occasional manual documents. | Matters for batches, reports, transcripts, or uploads. | 20k input / 1k output x 1k requests Summarization guide |
| Coding-agent workflow | Usually enough for personal coding tools. | Matters when your app runs agent turns for users. | 8k input / 4k output x 2k requests Coding-agent guide |