LLM Token Counter & API Cost Calculator

If you have ever shipped a feature that calls an LLM API and then watched your invoice triple overnight, you already know why token counting matters. A single misplaced system prompt or an untruncated document dump can turn a $50/month prototype into a $500 surprise. This tool lets you catch those problems before they hit production.

Paste any text — a system prompt, a few-shot example block, a full document — and the counter breaks down token counts and estimated costs for multiple models side by side. Supported providers include Claude (Opus, Sonnet, Haiku), GPT-4o, GPT-4o mini, and Gemini 1.5 Pro / Flash. Input and output pricing are shown separately so you can model real request/response pairs accurately.

A quick note on accuracy: this tool uses a character-based heuristic to approximate token counts rather than running each provider's exact tokenizer. For English text the estimate is typically within 5-10% of the actual count. CJK text (Chinese, Japanese, Korean) tends to tokenize less efficiently, so the tool applies a higher token-per-character ratio for those scripts. The numbers are good enough for budgeting and comparison, but if you need exact counts for billing reconciliation, use the official tokenizer libraries like tiktoken for OpenAI or Anthropic's token counting API.

You can pair this with the Word Counter to track both human-readable length and machine cost at the same time. If you are building prompts that include structured data, the JSON Formatter can help you minify payloads before pasting — fewer characters often means fewer tokens. For checking raw byte length of your prompts, the String Length Calculator is useful when you are close to context window limits.

Some practical tips for keeping costs down: enable prompt caching if your provider supports it — Anthropic and OpenAI both offer cached prompt reads at up to 90% off the standard input price, which adds up fast when your system prompt stays the same across requests. For non-latency-sensitive workloads, both providers offer batch APIs at roughly 50% off standard pricing. Beyond that, strip unnecessary whitespace from context documents and pick the cheapest model that meets your quality bar. Haiku and GPT-4o mini handle classification and extraction tasks well at a fraction of the cost of larger models. Keep an eye on each model's context window limit too — Claude supports up to 200K tokens, GPT-4o up to 128K, and Gemini up to 1M — exceeding these limits means you need to truncate or summarize input.

All processing runs entirely in your browser. Your text is never sent to any server — no API calls, no logging, no telemetry. Close the tab and the data is gone.

1. Paste or type your text into the Input area. This can be a system prompt, user message, document, or any content you plan to send to an LLM.
2. The token count and cost estimates update instantly as you type. No button press needed.
3. Review the model comparison table to see estimated token counts and per-request costs for each provider.
4. Toggle between Input and Output pricing modes to estimate costs for prompts vs. completions (output tokens are more expensive on most models).
5. Adjust the token count slider or type a number manually if you want to model a specific response length for the output side.
6. Click Copy or press Ctrl+Shift+C to copy the cost breakdown to your clipboard for sharing or documentation.
7. Use the Clear button to reset the input and start a new estimate.