Cost calculation (Polyglot)

Overview

Calculate the cost of an LLM inference call using FlatRateCostCalculator. Pricing is decoupled from usage — you define rates with InferencePricing, get token counts from the response's InferenceUsage, and the calculator produces a Cost value object with total and per-category breakdown.

Example

<?php
require 'examples/boot.php';

use Cognesy\Messages\Messages;
use Cognesy\Polyglot\Inference\Inference;
use Cognesy\Polyglot\Inference\Data\InferencePricing;
use Cognesy\Polyglot\Inference\Pricing\FlatRateCostCalculator;

// 1. Run an inference call
$prompt = 'Explain the observer pattern in one sentence.';
$response = Inference::using('openai')
    ->withMessages(Messages::fromString($prompt))
    ->withOptions(['max_tokens' => 100])
    ->response();

$usage = $response->usage();

echo "Prompt: {$prompt}\n";
echo "Response: {$response->content()}\n\n";
echo "Tokens — input: {$usage->inputTokens}, output: {$usage->outputTokens}\n\n";

// 2. Define pricing rates ($/1M tokens) and calculate cost
$calculator = new FlatRateCostCalculator();

$pricing = InferencePricing::fromArray([
    'input'  => 0.2,  // gpt-4.1-nano input
    'output' => 0.8,  // gpt-4.1-nano output
]);

$cost = $calculator->calculate($usage, $pricing);

echo "Cost breakdown:\n";
foreach ($cost->breakdown as $category => $amount) {
    if ($amount > 0) {
        printf("  %-12s \$%.6f\n", $category, $amount);
    }
}
echo "  Total:       {$cost->toString()}\n\n";

// 3. Compare across models using the same usage
$models = [
    'GPT-4.1-nano'      => ['input' => 0.2,  'output' => 0.8],
    'GPT-4.1-mini'      => ['input' => 0.4,  'output' => 1.6],
    'GPT-4.1'           => ['input' => 2.0,  'output' => 8.0],
    'Claude 4 Haiku'    => ['input' => 0.8,  'output' => 4.0],
    'Claude 4 Sonnet'   => ['input' => 3.0,  'output' => 15.0],
    'Gemini 2.5 Flash'  => ['input' => 0.15, 'output' => 0.60],
];

echo "Cost comparison ({$usage->inputTokens} in + {$usage->outputTokens} out tokens):\n";
foreach ($models as $model => $rates) {
    $modelCost = $calculator->calculate($usage, InferencePricing::fromArray($rates));
    printf("  %-20s %s\n", $model, $modelCost->toString());
}

// 4. Accumulate costs across multiple calls
$cost2 = $calculator->calculate($usage, InferencePricing::fromArray(['input' => 3.0, 'output' => 15.0]));
$totalCost = $cost->withAccumulated($cost2);
echo "\nAccumulated (nano + sonnet pricing): {$totalCost->toString()}\n";

assert($usage->inputTokens > 0);
assert($usage->outputTokens > 0);
assert($cost->total > 0);
assert($totalCost->total > $cost->total);
?>