Tool

LLM token counter and chatbot cost estimator

Token counts are a heuristic (roughly within 10 to 15 percent for English text); every model family uses its own tokenizer. Prices are example tiers, not quotes: enter your provider's current prices.

System prompt (0 tokens)

Typical user message (0 tokens, 0 characters total)

Output tokens per reply

History turns included

Tool schema tokens

Price tier (example)

Context window (tokens)

Input $ / 1M tokens

Output $ / 1M tokens

Requests per day

% of prefix cached

Cache discount %

Estimate

Where chatbot tokens go

Every request resends the system prompt, tool definitions and whatever history you include, so a long conversation gets more expensive with each turn. The cheapest wins are usually a shorter system prompt, a sliding window or summary of older turns, a cap on output length, and prompt caching for the stable prefix.

Rule of thumb for English: 1 token is about 4 characters or 0.75 words. Spanish and code use somewhat more tokens per word.
For exact counts use the tokenizer library or the token-count endpoint published by your model provider.