AI API pricing glossary

Short definitions of the words you will see on our price table, pricing guide and bill-cutting tips. No jargon beyond what you need to compare costs.

Token

A small piece of text the model counts — often part of a word. In English, about 1,000 tokens is roughly 750 words. Providers bill by how many tokens you send and how many the model writes.

Input tokens

The text you send to the model: your question, instructions, chat history and any pasted documents. Also called prompt tokens.

Output tokens

The text the model writes back. Output usually costs several times more per token than input, so long answers raise the bill fast.

Per 1M tokens

Prices are quoted per one million tokens (about 750,000 English words). A rate of $2 per 1M input tokens means that much text costs $2 to send in.

Context window

How much text the model can see at once (prompt plus reply), measured in tokens. A larger window lets you paste bigger documents; some providers charge more when the prompt is very long.

Cached input / prompt caching

A discount when you reuse the same long instructions or documents across many calls. Our table shows the standard (uncached) list rate; check the provider for cache prices.

Batch API

A cheaper way to run many non-urgent jobs together, often at about half the standard rate, with slower turnaround. Our table shows live/standard prices, not batch.

Long-context pricing

Higher per-token rates some providers charge when a single prompt is very long (for example over 200K or 272K tokens). Short-context list prices are what we show by default; long rates appear in each model’s note when we know them.

List / standard price

The published pay-as-you-go rate with no batch discount, no cached input and (unless noted) short-context prompts. That is what every row on TokenTally compares.

Capability tier

Our rough 1–4 judgment of how capable a model is for hard work (1 = light chat, 4 = frontier). It is not a benchmark. “Cheapest for this job” only ranks models at or above the tier the job needs.

Thinking / reasoning tokens

Extra tokens some models spend “thinking” before the final answer. Providers often bill those as output, so a short reply can still cost more than it looks.

Provider

The company that hosts the model API — for example OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, Groq or Together. The same open model can cost different amounts on different hosts.

New to the numbers? Read how AI pricing works. Ready to spend less? See how to cut your AI bill or example workloads. Or jump to models, providers, jobs and compares.

Definitions describe how TokenTally uses these terms. Provider billing details can differ — always check the official pricing page linked from each model.