LLM Token Calculator: How to Estimate and Cut Token Costs

Estimating LLM token usage before you call an API prevents budget surprises and helps you size prompts. A rough token calculator is simple: English text runs about 4 characters per token, or roughly 1.3-1.5 tokens per word.

The rule of thumb

• ~4 characters per token for English.
• ~1.3-1.5 tokens per word.
• ~100 tokens ≈ 75 words.
• Code and JSON tokenize denser than prose.

Calculating cost

1. Count input tokens: system prompt + context + user text.
2. Estimate output tokens from your desired answer length.
3. Apply the model's price: $/1M input and $/1M output.
4. Total = (input/1M × in-price) + (output/1M × out-price).

Worked example

Hidden token eaters

• Re-sending the same context on every call instead of caching.
• Huge system prompts restated per request.
• Output token ceilings set too high — models overgenerate.
• Agent loops that accumulate full conversation history.

Cutting token costs

• Trim context to only what the task needs.
• Use prompt caching for stable prefixes.
• Lower max output tokens.
• Route simple tasks to cheaper models.
• Summarize conversation history in long agents.

FAQ