LitellmPricingProvider

namespace Aaix\LaravelAiCosts\Support;

final class LitellmPricingProvider

Fetches and caches the LiteLLM community pricing database, then exposes per-model lookups in the package's normalised format.

getPricing()

/** @return array{input: float, output: float}|null */
public static function getPricing(string $model): ?array

Returns pricing in USD per 1M tokens for the given model key, or null if the model is not in the index. The model key can be either bare (claude-sonnet-4-6) or provider-scoped (anthropic/claude-sonnet-4-6) — the calculator tries the scoped form first.

clearCache()

public static function clearCache(): void

Forgets the ai-costs:litellm-index cache entry. The next getPricing() call will refetch from config('ai-costs.litellm.url').

Cache behaviour

• Cache key: ai-costs:litellm-index
• TTL: config('ai-costs.litellm.cache_ttl') seconds (default 86400 / 24h)
• Store: whatever Laravel's default cache store is configured to use
• HTTP timeout: 15 seconds
• On HTTP failure: returns an empty index (the calculator then throws InvalidArgumentException)

Index construction

Entries are skipped when:

• input_cost_per_token or output_cost_per_token is missing
• both costs are 0 (LiteLLM uses zeros as placeholders for free / unknown models)

Surviving entries are multiplied by 1_000_000 to convert from per-token to per-1M-token, matching the unit used by the models config array.