Function extractUsageAndCost

web/src/llm-api/fireworks.ts:203–252

(
  usage: Record<string, unknown> | undefined | null,
  model: string,
)

}

function extractUsageAndCost(
  usage: Record<string, unknown> | undefined | null,
  model: string,
): UsageData {
  if (!usage)
    return {
      inputTokens: 0,
      outputTokens: 0,
      cacheReadInputTokens: 0,
      reasoningTokens: 0,
      cost: 0,
    }
  const promptDetails = usage.prompt_tokens_details as
    | Record<string, unknown>
    | undefined
    | null
  const completionDetails = usage.completion_tokens_details as
    | Record<string, unknown>
    | undefined
    | null

  const inputTokens =
    typeof usage.prompt_tokens === 'number' ? usage.prompt_tokens : 0
  const outputTokens =
    typeof usage.completion_tokens === 'number' ? usage.completion_tokens : 0
  const cacheReadInputTokens =
    typeof promptDetails?.cached_tokens === 'number'
      ? promptDetails.cached_tokens
      : 0
  const reasoningTokens =
    typeof completionDetails?.reasoning_tokens === 'number'
      ? completionDetails.reasoning_tokens
      : 0

  // Fireworks doesn't return cost — compute from token counts and known pricing
  const pricing = getFireworksPricing(model)
  const nonCachedInputTokens = Math.max(0, inputTokens - cacheReadInputTokens)
  const cost =
    nonCachedInputTokens * pricing.inputCostPerToken +
    cacheReadInputTokens * pricing.cachedInputCostPerToken +
    outputTokens * pricing.outputCostPerToken

  return {
    inputTokens,
    outputTokens,
    cacheReadInputTokens,
    reasoningTokens,
    cost,
  }
}

export async function handleFireworksNonStream({
  body,