Token Pricing

Token pricing is the most natural fit for LLM-backed endpoints where the true cost scales with prompt size or model usage.

The tiktoken package is an optional peer dependency — only loaded when createTokenPricing() or countTokens() is called. Install it with npm install tiktoken if you use this feature.

OpenAI Example

import express from 'express';
import { createTokenPricing, createX402Server } from '@dexterai/x402/server';
 
const app = express();
const server = createX402Server({
  payTo: 'YourSolanaAddress...',
  network: 'solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp',
  facilitatorUrl: 'https://x402.dexter.cash',
});
 
const pricing = createTokenPricing({
  model: 'gpt-4o-mini',
});
 
app.use(express.json());
 
app.post('/api/chat', async (req, res) => {
  const { prompt, systemPrompt } = req.body;
  const paymentSig = req.headers['payment-signature'];
 
  if (!paymentSig) {
    // calculate() is async — tiktoken is lazy-loaded on first call
    const quote = await pricing.calculate(prompt, systemPrompt);
    const requirements = await server.buildRequirements({
      amountAtomic: quote.amountAtomic,
      resourceUrl: req.originalUrl,
      description: `${quote.model}: ${quote.inputTokens.toLocaleString()} input tokens`,
    });
 
    res.setHeader('PAYMENT-REQUIRED', server.encodeRequirements(requirements));
    res.setHeader('X-Quote-Hash', quote.quoteHash);
    return res.status(402).json({
      model: quote.model,
      inputTokens: quote.inputTokens,
      usdAmount: quote.usdAmount,
      tier: quote.tier,
    });
  }
 
  const quoteHash = req.headers['x-quote-hash'];
  // validateQuote() is also async
  if (!await pricing.validateQuote(prompt, quoteHash)) {
    return res.status(400).json({ error: 'Prompt changed, re-quote required' });
  }
 
  const result = await server.settlePayment(paymentSig);
  if (!result.success) {
    return res.status(402).json({ error: result.errorReason });
  }
 
  res.json({ ok: true, transaction: result.transaction });
});

All TokenPricing methods (calculate, validateQuote, countTokens) are async because tiktoken is lazy-loaded via dynamic import. This keeps the 5MB+ WASM binary out of your bundle if you don't use token pricing.

Custom Model Example

import { createTokenPricing } from '@dexterai/x402/server';
 
// Use custom rates for non-OpenAI models
const pricing = createTokenPricing({
  model: 'claude-3-sonnet',
  inputRate: 3.0,     // USD per 1M input tokens
  outputRate: 15.0,   // USD per 1M output tokens
  maxTokens: 4096,
});
 
// Custom tokenizer (sync or async)
const pricing = createTokenPricing({
  model: 'llama-3',
  inputRate: 0.5,
  tokenizer: (text) => myLlamaTokenizer.encode(text).length,
});

When This Is Better Than Generic Dynamic Pricing

Use token pricing when:

the endpoint wraps an LLM
model cost is the dominant pricing factor
buyers benefit from seeing model and token context in the quote

Use Dynamic Pricing when the cost model is not really about tokens.

Token Pricing

OpenAI Example

Custom Model Example

When This Is Better Than Generic Dynamic Pricing

On this page