Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Cost simulator
The cost simulator turns plain answers (people, chats, task size) into tokens using the planning figures below. They are our assumptions, not source evidence, and they never change a published price. Every figure can be overridden in the simulator’s token-level fields.
The Batch API pricing tier is offered in Automation and in the token-level fields, where every message is priced at the size you enter. Batch jobs run asynchronously, so chat and coding agents, which are interactive, are priced at standard rates only.
| Planning figure | Value |
|---|---|
| System prompt sent ahead of every turn (tokens) | 800 |
| One thing the reader types (tokens) | 80 |
| Share of the repeated prefix served from cache | 90% |
| Choices for “How long is a chat?” | Quick = 3 · Normal = 8 · Long thread = 20 |
| Choices for “What do people paste in?” | Nothing = 0 · A page = 1,500 · Long doc = 8,000 |
| Choices for “Typical answer length” | Short = 150 · Medium = 400 · Long = 1,000 |
| Light preset | Chats per person / workday: 4 · How long is a chat?: 3 · What do people paste in?: 0 · Typical answer length: 150 |
| Typical preset | Chats per person / workday: 10 · How long is a chat?: 8 · What do people paste in?: 0 · Typical answer length: 400 |
| Heavy preset | Chats per person / workday: 25 · How long is a chat?: 20 · What do people paste in?: 1,500 · Typical answer length: 1,000 |
| Planning figure | Value |
|---|---|
| System prompt plus tool definitions sent on every step (tokens) | 10,000 |
| Context added per step by file reads and tool output (tokens) | 2,000 |
| Context ceiling, where agents compact long sessions (tokens) | 180,000 |
| Share of the repeated prefix served from cache | 95% |
| Choices for “Typical task” | Small fix = 15 · Feature = 40 · Big refactor = 100 |
| Choices for “Reasoning effort” | Low = 300 · Medium = 700 · High = 1,500 |
| Light preset | Agent tasks per dev / workday: 2 · Typical task: 15 · Reasoning effort: 300 |
| Typical preset | Agent tasks per dev / workday: 6 · Typical task: 40 · Reasoning effort: 700 |
| Heavy preset | Agent tasks per dev / workday: 15 · Typical task: 100 · Reasoning effort: 1,500 |
| Planning figure | Value |
|---|---|
| Fixed instructions reused for every item (tokens) | 1,500 |
| Share of the fixed instructions served from cache | 95% |
| Choices for “Size of each item” | Short msg = 300 · Email / page = 1,500 · Long doc = 10,000 |
| Choices for “What comes back” | Label / JSON = 80 · Summary = 400 · Full draft = 1,200 |
| Choices for “Model calls per item” | Single call = 1 · Chain of 3 = 3 |
| Choices for “Runs on” | Weekdays = 22 · Every day = 30 |
| Choices for “Pricing tier” | Standard · Batch API |
| Light preset | Items per day: 100 · Size of each item: 300 · What comes back: 80 · Model calls per item: 1 |
| Typical preset | Items per day: 500 · Size of each item: 1,500 · What comes back: 400 · Model calls per item: 1 |
| Heavy preset | Items per day: 5,000 · Size of each item: 10,000 · What comes back: 1,200 · Model calls per item: 3 |