Function extractUsageAndCost

web/src/llm-api/siliconflow.ts:95–112

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

const SILICONFLOW_OUTPUT_COST_PER_TOKEN = 1.20 / 1_000_000

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

  const nonCachedInputTokens = Math.max(0, inputTokens - cacheReadInputTokens)
  const cost =
    nonCachedInputTokens * SILICONFLOW_INPUT_COST_PER_TOKEN +
    cacheReadInputTokens * SILICONFLOW_CACHED_INPUT_COST_PER_TOKEN +
    outputTokens * SILICONFLOW_OUTPUT_COST_PER_TOKEN

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

export async function handleSiliconFlowNonStream({
  body,