How to estimate tokens per call

Guide · sources linked inline · 2026-10-26

Input tokens are everything you send: the system prompt, tool/function definitions, retrieved documents, the conversation history and the user message. Output tokens are everything the model writes, including hidden reasoning tokens on reasoning models.

Practical method

1. Log 50–100 real requests from a prototype.
2. Read the token counts the API returns in each response (the usage field) instead of guessing.
3. Use the average for budgeting and the 90th percentile for worst-case planning.

Why counts differ between models

Each provider uses its own tokenizer, so the same text can be a different number of tokens. Anthropic notes that Claude 4.7 and newer use a tokenizer that can produce about 30% more tokens for the same text (source). Compare models on measured usage, not only on list price.

Chat history grows

In multi-turn chat, each turn re-sends the earlier conversation, so input tokens grow every turn. Budget with the average input size across a whole session, or use prompt caching for the repeated prefix.

Next step → Plug your numbers into the calculator

Prices last checked 2026-10-26 against official sources · how we verify