Glossary

Tokens, context window, model, API key, base URL, streaming, rate limit, credits, allowance, each in plain words with an example.

Short definitions of the words that appear across Halfbill and the tools you connect to it.

What you'll need

Nothing. Bookmark this page and come back when a word is unclear.

Token

The unit AI models read and write text in. A token is a word or a piece of a word; 1,000 tokens is roughly 750 English words. You pay per token, both for what you send and for what the model answers. Example: a two-paragraph question and a one-page answer is about 1,000 tokens.

Model

A specific AI "brain" you can talk to, such as Claude Sonnet 5 or GPT-5.5. Models differ in speed, price, and how well they handle hard tasks. Example: Haiku answers fast and cheap; Opus thinks harder and costs more; Fable 5.1 (claude-fable-5-1) is the strongest Anthropic model and the top choice for the hardest tasks.

Context window

How much text a model can keep in mind at once: your messages, the answers so far, and any files you attached. When a conversation grows past it, the oldest parts are dropped. Example: Claude models on Halfbill accept up to 200K tokens, roughly 150,000 words.

API

The "back door" that programs use to talk to Halfbill instead of a person typing in a chat window. Claude Code, Cursor, and your own apps all use it. You never see it directly.

API key

A long secret string, like a password, that you create in the dashboard and paste into a tool. It tells Halfbill "this tool may use my balance". Anyone who has the key can spend your credits, so treat it like a card number. Example: sk-..., shown once when you create it.

Base URL

The internet address a tool sends its requests to. Tools default to the vendor's address; you change it to Halfbill's. Example: https://api.halfbill.uk for Claude Code, https://api.halfbill.uk/v1 for Cursor and OpenAI-style tools.

Streaming

Showing the answer word by word as it is produced, instead of waiting for the whole answer. Most tools stream by default. It does not change the price.

Rate limit

A cap on how many requests you can send per minute. If you exceed it, requests are refused for a moment with a "rate limit" message, then work again. Normal use rarely reaches it. Example: a script that fires hundreds of requests at once.

Credits

Your prepaid balance in dollars. Each request takes its cost off the balance. Credits stay valid for 12 months after your last purchase. Example: $10 of credits buys $20 of usage at vendor prices, because Halfbill charges half the vendor price.

Allowance

The monthly amount of usage included in a Chat or Halfbill Code plan. It resets each month and does not carry over. Example: Code Starter gives you $100 of coding at vendor prices for $25/mo.

Priority models

Model ids ending in -max (for example claude-sonnet-5-max) are the same models with priority routing and better cache efficiency for coding agents. They are priced at 40% below list instead of 50%. Recommended only for Claude Code power users; everyone else should use the regular id.

Spend limit

A cap you can set on a single API key so a tool cannot spend more than that amount, in total or per period. When it is reached, that key stops working until you raise the limit; your other keys are unaffected.

Want the technical details?

The Overview and Rate limits pages give the exact numbers.

Next step

Troubleshooting →

How is this guide?

On this page