fix(types): normalize stream usage before cost calculation (#9977)

open-sse/utils/stream.ts

@@ -8,6 +8,7 @@ import {
   logUsage,
   addBufferToUsage,
   filterUsageForFormat,
+  normalizeUsage as normalizeTokenUsage,
 } from "./usageTracking.ts";
 import {
   parseSSELine,
@@ -1012,7 +1013,9 @@ export function createSSEStream(options: StreamOptions = {}) {
     // JSON.parse every SSE line will crash on `: x-omniroute-*` comment lines.
     if (!sseCommentsEnabled()) return;
 
-    const costUsd = finalUsage ? await calculateCost(provider, model, finalUsage) : 0;
+    const costUsd = finalUsage
+      ? await calculateCost(provider, model, normalizeTokenUsage(finalUsage))
+      : 0;
     const comment = buildOmniRouteSseMetadataComment({
       provider,
       model,

tests/unit/usage-extractor.test.ts

@@ -2,7 +2,19 @@ import test from "node:test";
 import assert from "node:assert/strict";
 
 const { extractUsageFromResponse } = await import("../../open-sse/handlers/usageExtractor.ts");
-const { extractUsage } = await import("../../open-sse/utils/usageTracking.ts");
+const { extractUsage, normalizeUsage } = await import("../../open-sse/utils/usageTracking.ts");
+
+test("normalizeUsage keeps only finite numeric fields for stream cost calculation", () => {
+  assert.deepEqual(
+    normalizeUsage({
+      input_tokens: "12",
+      output_tokens: 3,
+      total_tokens: Number.POSITIVE_INFINITY,
+      input_tokens_details: { cached_tokens: 2 },
+    }),
+    { input_tokens: 12, output_tokens: 3 }
+  );
+});
 
 test("extractUsageFromResponse reads OpenAI chat completion usage", () => {
   const usage = extractUsageFromResponse(