There are many ways to use an LLM. Most fall into two camps: a subscription interface (flat monthly fee, provider handles everything) or the API (pay per token, you build the experience). Same model underneath — very different relationship.
Subscription Interfaces — Pay Once, Use Freely
Most people’s first LLM experience is through a subscription product: you sign up, pay a flat monthly fee, and use the model through a polished interface. The provider handles everything — the interface, conversation history, file uploads, plugins.
These come in several forms:
Visit a website (claude.ai, chatgpt.com, gemini.google.com) and start typing. The most common entry point — nothing to install, works on any device.
Most providers have iOS and Android apps. Same subscription, native mobile experience — useful for on-the-go queries or voice input.
AI assistants built into code editors — GitHub Copilot, Cursor, Claude in VS Code. Used by developers who want AI suggestions without leaving their editor. Often covered by a separate subscription.
💡 IDE = Integrated Development Environment — a code editor like VS Code, Cursor, or JetBrains. AI plugins live directly inside your editor, letting you ask questions, generate code, and get explanations without switching windows.
Tools like Claude Code that run in a terminal. Developers use these to interact with an LLM directly from the command line — useful for tasks like reviewing code, running automated workflows, or querying the model from scripts.
💡 CLI = Command Line Interface — a text-based tool you run in your computer's terminal by typing commands, rather than clicking buttons in a graphical app. Common among developers and power users.
What these have in common: The provider manages everything. You pay a flat fee and get a curated experience. Usage limits may apply on lower tiers.
The API — Pay Per Token
The API is how developers and businesses talk to LLMs programmatically. There’s no interface: your code sends a request, the model responds in raw text, your code does something with it.
- Pay-on-demand — billed per million tokens (input + output), no monthly commitment
- No interface layer — you build whatever experience you want on top
- Scalable — process one request or one million, the pricing scales with you
- Same model as the subscription products — just a different delivery method
The API is the right tool when you’re:
- Automating — running the same task repeatedly without human involvement
- Building a product — embedding an LLM into your own app or workflow
- Integrating — connecting the LLM to your database, CRM, or email system
- Scaling — processing hundreds or thousands of requests programmatically
Providers heavily subsidize their subscription plans to attract users. At low to moderate usage, the API can cost significantly more than a flat subscription — sometimes 5–10×. For individuals and small teams, the subscription is often the better deal. The economics flip only when you’re building at scale.
Signs You’ve Outgrown the Subscription Interface
- You find yourself copying and pasting the same prompt repeatedly
- You want to process a list of things (emails, products, customers) through the LLM
- You need the LLM’s output to automatically feed into another system
- You want to build a tool that other people use
These are all API use cases.