LLM Token Costs: Why Your Prompt Might Cost 10x More Than You Think

If you're building with LLM APIs, you've probably wondered: how many tokens is this prompt actually using?

I built a free tool to answer that: LLM Token Counter

What it does

• Paste any text and instantly see the token count
• Compares costs across GPT-4o, GPT-3.5 Turbo, Claude 3 Haiku, and Gemini 1.5 Flash
• No login, no backend — runs entirely in your browser
• Uses the actual cl100k_base tokenizer (same as OpenAI) for accurate GPT counts

The cost gap is real

Consider a typical system prompt + user message (about 500 tokens):

Model	Cost per 1M tokens	Cost for 500 tokens
GPT-4o	$2.50	$0.00125
GPT-3.5 Turbo	$0.50	$0.00025
Claude 3 Haiku	$0.25	$0.000125
Gemini 1.5 Flash	$0.075	$0.0000375

That's a 33x cost difference between GPT-4o and Gemini 1.5 Flash for the same prompt.

Why client-side matters

Most token counters require you to send your prompt to a server. This one runs entirely in your browser using js-tiktoken — your prompts never leave your machine.

Try it

👉 https://code-two-delta.vercel.app

Source code: github.com/SolvoHQ/llm-token-counter

Would love feedback — what models or features would make this more useful for your workflow?