TokenMath.net — AI Token Counter & LLM Cost Calculator

TokenMath.net is a production-ready, privacy-first developer utility for real-time AI token counting, prompt caching economics modeling, multimodal calculations, and LLM API budget planning.

Key Features

• Universal Token Counter: Live token counts across GPT-6 (o200k_base), GPT-4 (cl100k_base), Claude 5 (calibrated BPE), Gemini 3 (SentencePiece), and DeepSeek/Llama.
• Cost Estimator & Budget Planner: Detailed input, cached read/write, output, and batch pricing with volume scaling.
• Price Comparison Matrix: Interactive 19-model price matrix with multi-column sorting and 1-click CSV Export.
• Multimodal Calculators:
  • Audio & Video: Gemini 1 fps video frame tokens (290 tok/s), Whisper STT per-minute rates, and OpenAI TTS.
  • Vision Calculator: High/low-res downscaling and 512×512 tiling calculations for OpenAI, Anthropic, and Google.
• Developer Tools:
  • Prompt Sanitizer & PII Redactor: 100% client-side redaction of API keys (sk-..., ghp_..., AWS, DB strings), emails, and IPs with whitespace minification.
  • Vector Embeddings & RAG Ingestion: Dimension sizes, chunking parameters, and vector database pricing.
  • GPU TCO Break-Even Solver: Rented H100/A100/L40S vs Cloud API cost inflection threshold.
• Model Directory (/models) & Dedicated Model Pages (/model/[slug]): 19 pre-rendered static deep-dive pages with Schema.org Product structured data.
• Developer Guide (/guide) & FAQ (/faq): 12-minute technical guide on caching architecture, batch discounts, and rate limits.
• TokenMath CLI Companion (npx @tokenmath/tokenmath): Zero-install terminal utility to calculate API invocation costs, compare models, and pipe prompt files into CI/CD budget guardrails.
• Automated Price Sync (/api/sync-prices): Programmatic endpoint ready for weekly GitHub Actions cron jobs.

TokenMath CLI Companion (@tokenmath/tokenmath)

Execute calculations and count tokens directly in your terminal with zero installation:

# Instant calculation with caching and explicit tokens
npx @tokenmath/tokenmath calculate --model gpt-5-6-sol --input 10000 --output 2000

# Pipe prompt files to count tokens
cat prompt.txt | npx @tokenmath/tokenmath count

# Head-to-head comparison
npx @tokenmath/tokenmath compare claude-sonnet-5 gpt-5-6-sol --input 10000 --output 2000

# Export master pricing registry as JSON
npx @tokenmath/tokenmath prices --json

Or install globally:

npm install -g @tokenmath/tokenmath
tokenmath calculate --model claude-sonnet-5 --prompt system.txt

Tech Stack

• Framework: Next.js 16 (App Router, Turbopack)
• Language: TypeScript (Strict Mode)
• Styling: Tailwind CSS, Lucide Icons
• Tokenizers: gpt-tokenizer (exact BPE) + calibrated statistical models
• Architecture: 100% Client-Side Calculations (Zero Prompts Transmitted or Stored)

Getting Started

1. Install Dependencies

npm install

2. Run Development Server

npm run dev

Open http://localhost:3000 in your browser.

3. Production Build

npm run build
npm run start -- -p 3030

Deployment Guide

Option 1: Vercel (Recommended)

1. Push your repository to GitHub:

   git init
   git add .
   git commit -m "feat: initial production release of TokenMath.net"
   git remote add origin https://github.com/<your-username>/tokenmath.git
   git push -u origin main

2. Go to vercel.com and click "Add New Project".
3. Import your tokenmath repository.
4. Next.js is auto-detected. Click Deploy.
5. Under Project Settings > Domains, connect tokenmath.net.

Option 2: Cloudflare Pages / Netlify

• Build command: npm run build
• Output directory: .next (or standalone)
• Node.js Version: >= 20.x

Post-Launch Monetization (Google AdSense)

1. Open public/ads.txt and replace pub-XXXXXXXXXXXXXXXX with your official Google AdSense publisher ID:

   google.com, pub-XXXXXXXXXXXXXXXX, DIRECT, f08c47fec0942fa0

2. When your AdSense account is approved, paste your client script in components/AdSlot.tsx.
3. Ad placements can be turned ON/OFF anytime via the hidden Admin Console at /admin.

License & Privacy

TokenMath.net is designed with a privacy-first philosophy. No user input, prompt text, uploaded documents, or calculated figures are ever sent to any remote server or third party.