Here's a strong first comment, tuned for Product Hunt's conventions (origin story + features + clear feedback ask). It runs ~290 words — the sweet spot for maker comments. Recommended first comment Hey Product Hunt! I built LLM Token Counter because I kept hitting the same annoying problem: every time I needed to estimate tokens for a prompt, I had to either spin up the OpenAI Playground or paste my prompt into some random tokenizer site I did not trust with my data. So I built one that: - Gives you exact tiktoken counts for GPT-4o, GPT-4.1, GPT-4, GPT-3.5, o1, and o3 using the real OpenAI tokenizer (cl100k_base + o200k_base) compiled to WASM - Provides accurate estimates for Claude, Llama 3, Gemini, and Mistral using each provider's published chars-per-token ratios (typically within 5-10% of real counts) - Visualizes your prompt live with color-coded chunks so you can see exactly how tokenization splits your text - Shows character, word, sentence, and line counts side by side - Includes a reference table of chars-per-token per model and built-in sample prompts (short, system prompt, code, long passage) Critically, everything runs 100% in your browser. No backend, no API key, no data leaves your device. Works offline after the first load. It is part of a free dev tools collection I maintain at singhajit.com/tools — 27 tools including JWT decoder, SQL formatter, regex tester, cron generator, snowflake decoder, and more. All client-side, all free, all without signup. What I would love feedback on: - Models I should add next (I am considering Mixtral 8x22B, DeepSeek V3, Qwen) - Are the chars-per-token estimates for non-OpenAI models accurate enough for your workflow, or do you need exact counts? - Any UX rough edges Thanks for checking it out! — Ajit

LLM Token Counter

Estimate prompt cost before you call the LLM API

Estimate prompt cost before you call the LLM API