@wix/evalforge-evaluator 0.206.0 → 0.208.0

package/build/index.mjs

@@ -10961,7 +10961,7 @@ var MODEL_PRICING = {
   "claude-opus-4-5": { input: 5, output: 25 },
   "claude-sonnet-4-5": { input: 3, output: 15 },
   "claude-haiku-4-5": { input: 1, output: 5 },
-  // Anthropic — Claude 4
+  // Anthropic — Claude 4.0 (retired; kept so historical runs still show cost)
   "claude-opus-4": { input: 15, output: 75 },
   "claude-sonnet-4": { input: 3, output: 15 },
   // OpenAI — GPT-5
@@ -11305,13 +11305,13 @@ async function executeWithAiSdk(context) {
       [text, steps, rawUsage] = await Promise.all([
         stream.text,
         stream.steps,
-        stream.usage
+        stream.totalUsage
       ]);
     } else {
       const result = await generateText(sdkParams);
       text = result.text;
       steps = result.steps;
-      rawUsage = result.usage;
+      rawUsage = result.totalUsage;
     }
     const durationMs = Date.now() - startTime;
     const usage = {
@@ -11322,7 +11322,6 @@ async function executeWithAiSdk(context) {
     const llmTrace = buildLLMTrace2(
       steps,
       durationMs,
-      usage,
       modelId,
       provider,
       startTime,
@@ -11471,7 +11470,7 @@ function findToolResultError(step) {
   }
   return null;
 }
-function buildLLMTrace2(steps, totalDurationMs, totalUsage, modelId, provider, executionStartMs, stepTimestamps) {
+function buildLLMTrace2(steps, totalDurationMs, modelId, provider, executionStartMs, stepTimestamps) {
   const traceSteps = steps.map((step, i) => {
     const stepFinishedAt = stepTimestamps[i] ?? executionStartMs;
     const stepStartedAt = i === 0 ? executionStartMs : stepTimestamps[i - 1] ?? executionStartMs;
@@ -11503,11 +11502,14 @@ function buildLLMTrace2(steps, totalDurationMs, totalUsage, modelId, provider, e
     };
   });
   const totalCostUsd = traceSteps.reduce((sum, s) => sum + s.costUsd, 0);
-  const finalTokens = {
-    prompt: totalUsage.inputTokens,
-    completion: totalUsage.outputTokens,
-    total: totalUsage.totalTokens
-  };
+  const finalTokens = traceSteps.reduce(
+    (acc, s) => ({
+      prompt: acc.prompt + s.tokenUsage.prompt,
+      completion: acc.completion + s.tokenUsage.completion,
+      total: acc.total + s.tokenUsage.total
+    }),
+    { prompt: 0, completion: 0, total: 0 }
+  );
   return {
     id: randomUUID3(),
     steps: traceSteps,