diff --git a/.changeset/eval-scorer-cost-telemetry.md b/.changeset/eval-scorer-cost-telemetry.md
new file mode 100644
index 000000000..9915238c1
--- /dev/null
+++ b/.changeset/eval-scorer-cost-telemetry.md
@@ -0,0 +1,11 @@
+---
+"@voltagent/core": patch
+---
+
+fix: emit LLM judge token and provider cost telemetry on eval scorer spans
+
+VoltAgent now records LLM judge model, token usage, cached tokens, reasoning tokens,
+and provider-reported cost details on live eval scorer spans.
+
+This makes scorer-side usage visible in observability backends and enables downstream
+cost aggregation to distinguish agent costs from eval scorer costs.
diff --git a/packages/core/src/agent/eval.ts b/packages/core/src/agent/eval.ts
index 0b0182e88..a82949cc6 100644
--- a/packages/core/src/agent/eval.ts
+++ b/packages/core/src/agent/eval.ts
@@ -56,6 +56,21 @@ interface ScoreMetrics {
   datasetMetadata?: ReturnType<typeof extractDatasetMetadataFromCombinedMetadata>;
 }
 
+interface JudgeTelemetry {
+  modelName?: string;
+  promptTokens?: number;
+  completionTokens?: number;
+  totalTokens?: number;
+  cachedTokens?: number;
+  reasoningTokens?: number;
+  providerCost?: {
+    cost?: number;
+    upstreamInferenceCost?: number;
+    upstreamInferenceInputCost?: number;
+    upstreamInferenceOutputCost?: number;
+  };
+}
+
 async function resolveScorerDescriptors(
   config: AgentEvalConfig,
   host: AgentEvalHost,
@@ -168,6 +183,46 @@ function createScorerSpanAttributes(
   if (metrics.datasetMetadata?.datasetItemHash) {
     attributes["eval.dataset.item_hash"] = metrics.datasetMetadata.datasetItemHash;
   }
+  const judgeTelemetry = extractJudgeTelemetry(metrics.combinedMetadata);
+  if (judgeTelemetry?.modelName) {
+    attributes["ai.model.name"] = judgeTelemetry.modelName;
+    const provider = judgeTelemetry.modelName.includes("/")
+      ? judgeTelemetry.modelName.split("/")[0]
+      : undefined;
+    if (provider) {
+      attributes["ai.model.provider"] = provider;
+    }
+  }
+  if (judgeTelemetry?.promptTokens !== undefined) {
+    attributes["usage.prompt_tokens"] = judgeTelemetry.promptTokens;
+  }
+  if (judgeTelemetry?.completionTokens !== undefined) {
+    attributes["usage.completion_tokens"] = judgeTelemetry.completionTokens;
+  }
+  if (judgeTelemetry?.totalTokens !== undefined) {
+    attributes["usage.total_tokens"] = judgeTelemetry.totalTokens;
+  }
+  if (judgeTelemetry?.cachedTokens !== undefined) {
+    attributes["usage.cached_tokens"] = judgeTelemetry.cachedTokens;
+  }
+  if (judgeTelemetry?.reasoningTokens !== undefined) {
+    attributes["usage.reasoning_tokens"] = judgeTelemetry.reasoningTokens;
+  }
+  if (judgeTelemetry?.providerCost?.cost !== undefined) {
+    attributes["usage.cost"] = judgeTelemetry.providerCost.cost;
+  }
+  if (judgeTelemetry?.providerCost?.upstreamInferenceCost !== undefined) {
+    attributes["usage.cost_details.upstream_inference_cost"] =
+      judgeTelemetry.providerCost.upstreamInferenceCost;
+  }
+  if (judgeTelemetry?.providerCost?.upstreamInferenceInputCost !== undefined) {
+    attributes["usage.cost_details.upstream_inference_input_cost"] =
+      judgeTelemetry.providerCost.upstreamInferenceInputCost;
+  }
+  if (judgeTelemetry?.providerCost?.upstreamInferenceOutputCost !== undefined) {
+    attributes["usage.cost_details.upstream_inference_output_cost"] =
+      judgeTelemetry.providerCost.upstreamInferenceOutputCost;
+  }
   if (storagePayload.userId) {
     attributes["user.id"] = storagePayload.userId;
   }
@@ -1353,6 +1408,89 @@ function extractErrorMessage(error: unknown): string {
   }
 }
 
+function extractJudgeTelemetry(
+  metadata: Record<string, unknown> | null | undefined,
+): JudgeTelemetry | undefined {
+  const record = isPlainRecord(metadata) ? (metadata as Record<string, unknown>) : undefined;
+  if (!record) {
+    return undefined;
+  }
+
+  const sources: Array<Record<string, unknown> | undefined> = [];
+  if (isPlainRecord(record.voltAgent)) {
+    sources.push(record.voltAgent as Record<string, unknown>);
+  }
+  if (isPlainRecord(record.scorer)) {
+    sources.push(record.scorer as Record<string, unknown>);
+  }
+  if (isPlainRecord(record.payload)) {
+    sources.push(record.payload as Record<string, unknown>);
+  }
+
+  for (const source of sources) {
+    const judge = isPlainRecord(source?.judge)
+      ? (source?.judge as Record<string, unknown>)
+      : undefined;
+    if (!judge) {
+      continue;
+    }
+
+    const usage = isPlainRecord(judge.usage) ? (judge.usage as Record<string, unknown>) : undefined;
+    const providerCost = isPlainRecord(judge.providerCost)
+      ? (judge.providerCost as Record<string, unknown>)
+      : undefined;
+
+    const telemetry: JudgeTelemetry = {
+      modelName: readString(judge.model),
+      promptTokens: readNumber(usage?.promptTokens),
+      completionTokens: readNumber(usage?.completionTokens),
+      totalTokens: readNumber(usage?.totalTokens),
+      cachedTokens: readNumber(usage?.cachedInputTokens ?? usage?.cachedTokens),
+      reasoningTokens: readNumber(usage?.reasoningTokens),
+      providerCost: providerCost
+        ? {
+            cost: readNumber(providerCost.cost),
+            upstreamInferenceCost: readNumber(providerCost.upstreamInferenceCost),
+            upstreamInferenceInputCost: readNumber(providerCost.upstreamInferenceInputCost),
+            upstreamInferenceOutputCost: readNumber(providerCost.upstreamInferenceOutputCost),
+          }
+        : undefined,
+    };
+
+    if (
+      telemetry.modelName ||
+      telemetry.promptTokens !== undefined ||
+      telemetry.completionTokens !== undefined ||
+      telemetry.totalTokens !== undefined ||
+      telemetry.cachedTokens !== undefined ||
+      telemetry.reasoningTokens !== undefined ||
+      telemetry.providerCost?.cost !== undefined ||
+      telemetry.providerCost?.upstreamInferenceCost !== undefined ||
+      telemetry.providerCost?.upstreamInferenceInputCost !== undefined ||
+      telemetry.providerCost?.upstreamInferenceOutputCost !== undefined
+    ) {
+      return telemetry;
+    }
+  }
+
+  return undefined;
+}
+
+function readString(value: unknown): string | undefined {
+  return typeof value === "string" && value.length > 0 ? value : undefined;
+}
+
+function readNumber(value: unknown): number | undefined {
+  if (typeof value === "number") {
+    return Number.isFinite(value) ? value : undefined;
+  }
+  if (typeof value === "string") {
+    const parsed = Number(value);
+    return Number.isFinite(parsed) ? parsed : undefined;
+  }
+  return undefined;
+}
+
 async function invokeEvalResultCallback(
   host: AgentEvalHost,
   config: AgentEvalScorerConfig,
diff --git a/packages/core/src/eval/llm/create-judge-scorer.ts b/packages/core/src/eval/llm/create-judge-scorer.ts
index 1aee60242..a3485f280 100644
--- a/packages/core/src/eval/llm/create-judge-scorer.ts
+++ b/packages/core/src/eval/llm/create-judge-scorer.ts
@@ -2,10 +2,19 @@ import { safeStringify } from "@voltagent/internal/utils";
 import type { LanguageModel } from "ai";
 import { generateText } from "ai";
 
+import { convertUsage } from "../../utils/usage-converter";
 import type { LocalScorerDefinition } from "../runtime";
 
 type DefaultPayload = Record<string, unknown>;
 
+type OpenRouterUsageCost = {
+  cost?: number;
+  isByok?: boolean;
+  upstreamInferenceCost?: number;
+  upstreamInferenceInputCost?: number;
+  upstreamInferenceOutputCost?: number;
+};
+
 export interface LlmJudgeScorerParams extends Record<string, unknown> {
   /** Optional criteria appended to the default judging instructions. */
   criteria?: string;
@@ -49,13 +58,16 @@ export function createLLMJudgeScorer<Payload extends DefaultPayload = DefaultPay
       const criteria = params.criteria ? params.criteria.trim() : "";
 
       const prompt = buildPrompt({ instructions, criteria, question, answer });
+      const judgeModel = resolveJudgeModelId(model);
 
       try {
-        const { text } = await generateText({
+        const { text, usage, providerMetadata } = await generateText({
           model,
           prompt,
           maxOutputTokens,
         });
+        const normalizedUsage = convertUsage(usage);
+        const providerCost = extractOpenRouterUsageCost(providerMetadata);
 
         const parsed = parseJudgeResponse(text);
         if (!parsed) {
@@ -66,6 +78,11 @@ export function createLLMJudgeScorer<Payload extends DefaultPayload = DefaultPay
               raw: text.trim(),
               voltAgent: {
                 scorer: scorerId,
+                judge: buildJudgeTelemetry({
+                  judgeModel,
+                  usage: normalizedUsage,
+                  providerCost,
+                }),
               },
             },
             error: new Error("Judge response was not valid JSON"),
@@ -80,6 +97,11 @@ export function createLLMJudgeScorer<Payload extends DefaultPayload = DefaultPay
             raw: text.trim(),
             voltAgent: {
               scorer: scorerId,
+              judge: buildJudgeTelemetry({
+                judgeModel,
+                usage: normalizedUsage,
+                providerCost,
+              }),
             },
           },
         };
@@ -90,6 +112,11 @@ export function createLLMJudgeScorer<Payload extends DefaultPayload = DefaultPay
           metadata: {
             voltAgent: {
               scorer: scorerId,
+              judge: buildJudgeTelemetry({
+                judgeModel,
+                usage: extractUsageFromError(error),
+                providerCost: extractProviderCostFromError(error),
+              }),
             },
           },
           error,
@@ -178,3 +205,112 @@ function stringify(value: unknown): string {
     return String(value);
   }
 }
+
+function buildJudgeTelemetry(input: {
+  judgeModel: string;
+  usage?: ReturnType<typeof convertUsage>;
+  providerCost?: OpenRouterUsageCost;
+}): Record<string, unknown> {
+  const judge: Record<string, unknown> = {
+    model: input.judgeModel,
+  };
+
+  if (input.usage) {
+    judge.usage = input.usage;
+  }
+
+  if (input.providerCost) {
+    judge.providerCost = input.providerCost;
+  }
+
+  return judge;
+}
+
+function extractOpenRouterUsageCost(providerMetadata: unknown): OpenRouterUsageCost | undefined {
+  if (!isRecord(providerMetadata)) {
+    return undefined;
+  }
+
+  const openRouterMetadata = isRecord(providerMetadata.openrouter)
+    ? providerMetadata.openrouter
+    : undefined;
+  const usage = isRecord(openRouterMetadata?.usage) ? openRouterMetadata.usage : undefined;
+  const costDetails = isRecord(usage?.costDetails) ? usage.costDetails : undefined;
+
+  if (!usage) {
+    return undefined;
+  }
+
+  const result: OpenRouterUsageCost = {};
+  const cost = toFiniteNumber(usage.cost);
+  if (cost !== undefined) {
+    result.cost = cost;
+  }
+
+  if (typeof usage.isByok === "boolean") {
+    result.isByok = usage.isByok;
+  }
+
+  const upstreamInferenceCost = toFiniteNumber(costDetails?.upstreamInferenceCost);
+  if (upstreamInferenceCost !== undefined) {
+    result.upstreamInferenceCost = upstreamInferenceCost;
+  }
+
+  const upstreamInferenceInputCost = toFiniteNumber(costDetails?.upstreamInferenceInputCost);
+  if (upstreamInferenceInputCost !== undefined) {
+    result.upstreamInferenceInputCost = upstreamInferenceInputCost;
+  }
+
+  const upstreamInferenceOutputCost = toFiniteNumber(costDetails?.upstreamInferenceOutputCost);
+  if (upstreamInferenceOutputCost !== undefined) {
+    result.upstreamInferenceOutputCost = upstreamInferenceOutputCost;
+  }
+
+  return Object.keys(result).length > 0 ? result : undefined;
+}
+
+function extractUsageFromError(error: unknown): ReturnType<typeof convertUsage> | undefined {
+  if (!isRecord(error)) {
+    return undefined;
+  }
+
+  const usage = error.usage;
+  return usage ? convertUsage(usage as any) : undefined;
+}
+
+function extractProviderCostFromError(error: unknown): OpenRouterUsageCost | undefined {
+  if (!isRecord(error)) {
+    return undefined;
+  }
+
+  return extractOpenRouterUsageCost(error.providerMetadata);
+}
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+
+function toFiniteNumber(value: unknown): number | undefined {
+  if (typeof value === "number") {
+    return Number.isFinite(value) ? value : undefined;
+  }
+
+  if (typeof value === "string") {
+    const parsed = Number(value);
+    return Number.isFinite(parsed) ? parsed : undefined;
+  }
+
+  return undefined;
+}
+
+function resolveJudgeModelId(model: LanguageModel): string {
+  if (typeof model === "string") {
+    return model;
+  }
+
+  if ("modelId" in model && typeof model.modelId === "string" && model.modelId.length > 0) {
+    return model.modelId;
+  }
+
+  return "unknown";
+}