@browserbasehq/stagehand 3.0.2 → 3.0.3-alpha-c76ade009ef81208accae6475ec4707d3906e566

package/dist/index.d.ts

@@ -1132,6 +1132,8 @@ interface AgentResult {
     usage?: {
         input_tokens: number;
         output_tokens: number;
+        reasoning_tokens?: number;
+        cached_input_tokens?: number;
         inference_time_ms: number;
     };
 }
@@ -1346,18 +1348,28 @@ declare enum V3FunctionName {
 interface StagehandMetrics {
     actPromptTokens: number;
     actCompletionTokens: number;
+    actReasoningTokens: number;
+    actCachedInputTokens: number;
     actInferenceTimeMs: number;
     extractPromptTokens: number;
     extractCompletionTokens: number;
+    extractReasoningTokens: number;
+    extractCachedInputTokens: number;
     extractInferenceTimeMs: number;
     observePromptTokens: number;
     observeCompletionTokens: number;
+    observeReasoningTokens: number;
+    observeCachedInputTokens: number;
     observeInferenceTimeMs: number;
     agentPromptTokens: number;
     agentCompletionTokens: number;
+    agentReasoningTokens: number;
+    agentCachedInputTokens: number;
     agentInferenceTimeMs: number;
     totalPromptTokens: number;
     totalCompletionTokens: number;
+    totalReasoningTokens: number;
+    totalCachedInputTokens: number;
     totalInferenceTimeMs: number;
 }
 
@@ -1898,7 +1910,7 @@ declare class V3 {
      */
     get history(): Promise<ReadonlyArray<HistoryEntry>>;
     addToHistory(method: HistoryEntry["method"], parameters: unknown, result?: unknown): void;
-    updateMetrics(functionName: V3FunctionName, promptTokens: number, completionTokens: number, inferenceTimeMs: number): void;
+    updateMetrics(functionName: V3FunctionName, promptTokens: number, completionTokens: number, reasoningTokens: number, cachedInputTokens: number, inferenceTimeMs: number): void;
     private updateTotalMetrics;
     private _immediateShutdown;
     private static _installProcessGuards;

package/dist/index.js

@@ -85,7 +85,7 @@ var __async = (__this, __arguments, generator) => {
 var STAGEHAND_VERSION;
 var init_version = __esm({
   "lib/version.ts"() {
-    STAGEHAND_VERSION = "3.0.2";
+    STAGEHAND_VERSION = "3.0.3-alpha-c76ade009ef81208accae6475ec4707d3906e566";
   }
 });
 
@@ -8983,6 +8983,8 @@ var AgentCache = class {
         result.usage = {
           input_tokens: 0,
           output_tokens: 0,
+          reasoning_tokens: 0,
+          cached_input_tokens: 0,
           inference_time_ms: 0
         };
         result.metadata = __spreadProps(__spreadValues({}, (_b = result.metadata) != null ? _b : {}), {
@@ -9457,7 +9459,7 @@ function extract(_0) {
     userProvidedInstructions,
     logInferenceToFile = false
   }) {
-    var _a, _b, _c, _d, _e, _f, _g, _h;
+    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p;
     const metadataSchema = import_v32.z.object({
       progress: import_v32.z.string().describe(
         "progress of what has been extracted so far, as concise as possible"
@@ -9521,6 +9523,8 @@ function extract(_0) {
         LLM_output_file: extractResponseFile,
         prompt_tokens: (_a = extractUsage == null ? void 0 : extractUsage.prompt_tokens) != null ? _a : 0,
         completion_tokens: (_b = extractUsage == null ? void 0 : extractUsage.completion_tokens) != null ? _b : 0,
+        reasoning_tokens: (_c = extractUsage == null ? void 0 : extractUsage.reasoning_tokens) != null ? _c : 0,
+        cached_input_tokens: (_d = extractUsage == null ? void 0 : extractUsage.cached_input_tokens) != null ? _d : 0,
         inference_time_ms: extractEndTime - extractStartTime
       });
     }
@@ -9582,14 +9586,18 @@ function extract(_0) {
         timestamp: metadataCallTimestamp,
         LLM_input_file: metadataCallFile,
         LLM_output_file: metadataResponseFile,
-        prompt_tokens: (_c = metadataResponseUsage == null ? void 0 : metadataResponseUsage.prompt_tokens) != null ? _c : 0,
-        completion_tokens: (_d = metadataResponseUsage == null ? void 0 : metadataResponseUsage.completion_tokens) != null ? _d : 0,
+        prompt_tokens: (_e = metadataResponseUsage == null ? void 0 : metadataResponseUsage.prompt_tokens) != null ? _e : 0,
+        completion_tokens: (_f = metadataResponseUsage == null ? void 0 : metadataResponseUsage.completion_tokens) != null ? _f : 0,
+        reasoning_tokens: (_g = metadataResponseUsage == null ? void 0 : metadataResponseUsage.reasoning_tokens) != null ? _g : 0,
+        cached_input_tokens: (_h = metadataResponseUsage == null ? void 0 : metadataResponseUsage.cached_input_tokens) != null ? _h : 0,
         inference_time_ms: metadataEndTime - metadataStartTime
       });
     }
-    const totalPromptTokens = ((_e = extractUsage == null ? void 0 : extractUsage.prompt_tokens) != null ? _e : 0) + ((_f = metadataResponseUsage == null ? void 0 : metadataResponseUsage.prompt_tokens) != null ? _f : 0);
-    const totalCompletionTokens = ((_g = extractUsage == null ? void 0 : extractUsage.completion_tokens) != null ? _g : 0) + ((_h = metadataResponseUsage == null ? void 0 : metadataResponseUsage.completion_tokens) != null ? _h : 0);
+    const totalPromptTokens = ((_i = extractUsage == null ? void 0 : extractUsage.prompt_tokens) != null ? _i : 0) + ((_j = metadataResponseUsage == null ? void 0 : metadataResponseUsage.prompt_tokens) != null ? _j : 0);
+    const totalCompletionTokens = ((_k = extractUsage == null ? void 0 : extractUsage.completion_tokens) != null ? _k : 0) + ((_l = metadataResponseUsage == null ? void 0 : metadataResponseUsage.completion_tokens) != null ? _l : 0);
     const totalInferenceTimeMs = extractEndTime - extractStartTime + (metadataEndTime - metadataStartTime);
+    const totalReasoningTokens = ((_m = extractUsage == null ? void 0 : extractUsage.reasoning_tokens) != null ? _m : 0) + ((_n = metadataResponseUsage == null ? void 0 : metadataResponseUsage.reasoning_tokens) != null ? _n : 0);
+    const totalCachedInputTokens = ((_o = extractUsage == null ? void 0 : extractUsage.cached_input_tokens) != null ? _o : 0) + ((_p = metadataResponseUsage == null ? void 0 : metadataResponseUsage.cached_input_tokens) != null ? _p : 0);
     return __spreadProps(__spreadValues({}, extractedData), {
       metadata: {
         completed: metadataResponseCompleted,
@@ -9597,6 +9605,8 @@ function extract(_0) {
       },
       prompt_tokens: totalPromptTokens,
       completion_tokens: totalCompletionTokens,
+      reasoning_tokens: totalReasoningTokens,
+      cached_input_tokens: totalCachedInputTokens,
       inference_time_ms: totalInferenceTimeMs
     });
   });
@@ -9610,7 +9620,7 @@ function observe(_0) {
     logger,
     logInferenceToFile = false
   }) {
-    var _a, _b, _c, _d;
+    var _a, _b, _c, _d, _e, _f;
     const isGPT5 = llmClient.modelName.includes("gpt-5");
     const observeSchema = import_v32.z.object({
       elements: import_v32.z.array(
@@ -9670,6 +9680,8 @@ function observe(_0) {
     const { data: observeData, usage: observeUsage } = rawResponse;
     const promptTokens = (_a = observeUsage == null ? void 0 : observeUsage.prompt_tokens) != null ? _a : 0;
     const completionTokens = (_b = observeUsage == null ? void 0 : observeUsage.completion_tokens) != null ? _b : 0;
+    const reasoningTokens = (_c = observeUsage == null ? void 0 : observeUsage.reasoning_tokens) != null ? _c : 0;
+    const cachedInputTokens = (_d = observeUsage == null ? void 0 : observeUsage.cached_input_tokens) != null ? _d : 0;
     let responseFile = "";
     if (logInferenceToFile) {
       const { fileName: responseFileName } = writeTimestampedTxtFile(
@@ -9688,10 +9700,12 @@ function observe(_0) {
         LLM_output_file: responseFile,
         prompt_tokens: promptTokens,
         completion_tokens: completionTokens,
+        reasoning_tokens: reasoningTokens,
+        cached_input_tokens: cachedInputTokens,
         inference_time_ms: usageTimeMs
       });
     }
-    const parsedElements = (_d = (_c = observeData.elements) == null ? void 0 : _c.map((el) => {
+    const parsedElements = (_f = (_e = observeData.elements) == null ? void 0 : _e.map((el) => {
       const base = {
         elementId: el.elementId,
         description: String(el.description),
@@ -9699,11 +9713,13 @@ function observe(_0) {
         arguments: el.arguments
       };
       return base;
-    })) != null ? _d : [];
+    })) != null ? _f : [];
     return {
       elements: parsedElements,
       prompt_tokens: promptTokens,
       completion_tokens: completionTokens,
+      reasoning_tokens: reasoningTokens,
+      cached_input_tokens: cachedInputTokens,
       inference_time_ms: usageTimeMs
     };
   });
@@ -9717,7 +9733,7 @@ function act(_0) {
     logger,
     logInferenceToFile = false
   }) {
-    var _a, _b;
+    var _a, _b, _c, _d;
     const isGPT5 = llmClient.modelName.includes("gpt-5");
     const actSchema = import_v32.z.object({
       elementId: import_v32.z.string().describe(
@@ -9772,6 +9788,8 @@ function act(_0) {
     const { data: actData, usage: actUsage } = rawResponse;
     const promptTokens = (_a = actUsage == null ? void 0 : actUsage.prompt_tokens) != null ? _a : 0;
     const completionTokens = (_b = actUsage == null ? void 0 : actUsage.completion_tokens) != null ? _b : 0;
+    const reasoningTokens = (_c = actUsage == null ? void 0 : actUsage.reasoning_tokens) != null ? _c : 0;
+    const cachedInputTokens = (_d = actUsage == null ? void 0 : actUsage.cached_input_tokens) != null ? _d : 0;
     let responseFile = "";
     if (logInferenceToFile) {
       const { fileName: responseFileName } = writeTimestampedTxtFile(
@@ -9790,6 +9808,8 @@ function act(_0) {
         LLM_output_file: responseFile,
         prompt_tokens: promptTokens,
         completion_tokens: completionTokens,
+        reasoning_tokens: reasoningTokens,
+        cached_input_tokens: cachedInputTokens,
         inference_time_ms: usageTimeMs
       });
     }
@@ -9803,6 +9823,8 @@ function act(_0) {
       element: parsedElement,
       prompt_tokens: promptTokens,
       completion_tokens: completionTokens,
+      reasoning_tokens: reasoningTokens,
+      cached_input_tokens: cachedInputTokens,
       inference_time_ms: usageTimeMs,
       twoStep: actData.twoStep
     };
@@ -10493,7 +10515,7 @@ var ActHandler = class {
       const { instruction, page, variables, timeout, model } = params;
       const llmClient = this.resolveLlmClient(model);
       const doObserveAndAct = () => __async(this, null, function* () {
-        var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j;
+        var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n;
         yield waitForDomNetworkQuiet(
           page.mainFrame(),
           this.defaultDomSettleTimeoutMs
@@ -10518,12 +10540,16 @@ var ActHandler = class {
         });
         const actPromptTokens = (_b = actInferenceResponse.prompt_tokens) != null ? _b : 0;
         const actCompletionTokens = (_c = actInferenceResponse.completion_tokens) != null ? _c : 0;
-        const actInferenceTimeMs = (_d = actInferenceResponse.inference_time_ms) != null ? _d : 0;
-        (_e = this.onMetrics) == null ? void 0 : _e.call(
+        const actReasoningTokens = (_d = actInferenceResponse.reasoning_tokens) != null ? _d : 0;
+        const actCachedInputTokens = (_e = actInferenceResponse.cached_input_tokens) != null ? _e : 0;
+        const actInferenceTimeMs = (_f = actInferenceResponse.inference_time_ms) != null ? _f : 0;
+        (_g = this.onMetrics) == null ? void 0 : _g.call(
           this,
           "ACT" /* ACT */,
           actPromptTokens,
           actCompletionTokens,
+          actReasoningTokens,
+          actCachedInputTokens,
           actInferenceTimeMs
         );
         const raw = actInferenceResponse.element;
@@ -10588,7 +10614,7 @@ var ActHandler = class {
         if (!diffedTree.trim()) {
           diffedTree = combinedTree2;
         }
-        const combinedXpathMap2 = (_f = secondSnapshot.combinedXpathMap) != null ? _f : {};
+        const combinedXpathMap2 = (_h = secondSnapshot.combinedXpathMap) != null ? _h : {};
         const previousAction = `method: ${chosen.method}, description: ${chosen.description}, arguments: ${chosen.arguments}`;
         const stepTwoInstructions = buildStepTwoPrompt(
           instruction,
@@ -10606,12 +10632,14 @@ var ActHandler = class {
           logger: v3Logger,
           logInferenceToFile: this.logInferenceToFile
         });
-        (_j = this.onMetrics) == null ? void 0 : _j.call(
+        (_n = this.onMetrics) == null ? void 0 : _n.call(
           this,
           "ACT" /* ACT */,
-          (_g = action2.prompt_tokens) != null ? _g : 0,
-          (_h = action2.completion_tokens) != null ? _h : 0,
-          (_i = action2.inference_time_ms) != null ? _i : 0
+          (_i = action2.prompt_tokens) != null ? _i : 0,
+          (_j = action2.completion_tokens) != null ? _j : 0,
+          (_k = action2.reasoning_tokens) != null ? _k : 0,
+          (_l = action2.cached_input_tokens) != null ? _l : 0,
+          (_m = action2.inference_time_ms) != null ? _m : 0
         );
         const raw2 = action2.element;
         const result2 = (() => {
@@ -10679,7 +10707,7 @@ var ActHandler = class {
   }
   actFromObserveResult(action, page, domSettleTimeoutMs, llmClientOverride) {
     return __async(this, null, function* () {
-      var _a, _b, _c, _d, _e, _f;
+      var _a, _b, _c, _d, _e, _f, _g, _h;
       const settleTimeout = domSettleTimeoutMs != null ? domSettleTimeoutMs : this.defaultDomSettleTimeoutMs;
       const effectiveClient = llmClientOverride != null ? llmClientOverride : this.llmClient;
       const method = (_a = action.method) == null ? void 0 : _a.trim();
@@ -10756,12 +10784,14 @@ var ActHandler = class {
               logger: v3Logger,
               logInferenceToFile: this.logInferenceToFile
             });
-            (_e = this.onMetrics) == null ? void 0 : _e.call(
+            (_g = this.onMetrics) == null ? void 0 : _g.call(
               this,
               "ACT" /* ACT */,
               (_b = actInferenceResponse.prompt_tokens) != null ? _b : 0,
               (_c = actInferenceResponse.completion_tokens) != null ? _c : 0,
-              (_d = actInferenceResponse.inference_time_ms) != null ? _d : 0
+              (_d = actInferenceResponse.reasoning_tokens) != null ? _d : 0,
+              (_e = actInferenceResponse.cached_input_tokens) != null ? _e : 0,
+              (_f = actInferenceResponse.inference_time_ms) != null ? _f : 0
             );
             const fallback = actInferenceResponse.element;
             if (!fallback) {
@@ -10775,7 +10805,7 @@ var ActHandler = class {
             let newSelector = action.selector;
             if (typeof fallback.elementId === "string") {
               const enc = fallback.elementId;
-              const rawXp = ((_f = snapshot.combinedXpathMap) != null ? _f : {})[enc];
+              const rawXp = ((_h = snapshot.combinedXpathMap) != null ? _h : {})[enc];
               const trimmed = trimTrailingTextNode(rawXp);
               if (trimmed) newSelector = `xpath=${trimmed}`;
             }
@@ -10894,11 +10924,15 @@ var ExtractHandler = class {
           metadata: { completed },
           prompt_tokens,
           completion_tokens,
+          reasoning_tokens = 0,
+          cached_input_tokens = 0,
           inference_time_ms
         } = _d, rest = __objRest(_d, [
           "metadata",
           "prompt_tokens",
           "completion_tokens",
+          "reasoning_tokens",
+          "cached_input_tokens",
           "inference_time_ms"
         ]);
         let output = rest;
@@ -10923,6 +10957,8 @@ var ExtractHandler = class {
           "EXTRACT" /* EXTRACT */,
           prompt_tokens,
           completion_tokens,
+          reasoning_tokens,
+          cached_input_tokens,
           inference_time_ms
         );
         const idToUrl = combinedUrlMap != null ? combinedUrlMap : {};
@@ -11007,6 +11043,8 @@ var ObserveHandler = class {
         const {
           prompt_tokens = 0,
           completion_tokens = 0,
+          reasoning_tokens = 0,
+          cached_input_tokens = 0,
           inference_time_ms = 0
         } = observationResponse;
         (_c = this.onMetrics) == null ? void 0 : _c.call(
@@ -11014,6 +11052,8 @@ var ObserveHandler = class {
           "OBSERVE" /* OBSERVE */,
           prompt_tokens,
           completion_tokens,
+          reasoning_tokens,
+          cached_input_tokens,
           inference_time_ms
         );
         const elementsWithSelectors = (yield Promise.all(
@@ -11724,6 +11764,8 @@ var V3AgentHandler = class {
             "AGENT" /* AGENT */,
             result.usage.inputTokens || 0,
             result.usage.outputTokens || 0,
+            result.usage.reasoningTokens || 0,
+            result.usage.cachedInputTokens || 0,
             inferenceTimeMs
           );
         }
@@ -11735,6 +11777,8 @@ var V3AgentHandler = class {
           usage: result.usage ? {
             input_tokens: result.usage.inputTokens || 0,
             output_tokens: result.usage.outputTokens || 0,
+            reasoning_tokens: result.usage.reasoningTokens || 0,
+            cached_input_tokens: result.usage.cachedInputTokens || 0,
             inference_time_ms: inferenceTimeMs
           } : void 0
         };
@@ -14091,6 +14135,7 @@ var V3CuaAgentHandler = class {
   }
   execute(optionsOrInstruction) {
     return __async(this, null, function* () {
+      var _a, _b;
       const options = typeof optionsOrInstruction === "string" ? { instruction: optionsOrInstruction } : optionsOrInstruction;
       this.highlightCursor = options.highlightCursor !== false;
       const page = yield this.v3.context.awaitActivePage();
@@ -14123,6 +14168,8 @@ var V3CuaAgentHandler = class {
           "AGENT" /* AGENT */,
           result.usage.input_tokens,
           result.usage.output_tokens,
+          (_a = result.usage.reasoning_tokens) != null ? _a : 0,
+          (_b = result.usage.cached_input_tokens) != null ? _b : 0,
           inferenceTimeMs
         );
       }
@@ -15672,7 +15719,7 @@ var AISdkClient = class extends LLMClient {
     return __async(this, arguments, function* ({
       options
     }) {
-      var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o;
+      var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s;
       (_a = this.logger) == null ? void 0 : _a.call(this, {
         category: "aisdk",
         message: "creating chat completion",
@@ -15803,10 +15850,12 @@ var AISdkClient = class extends LLMClient {
           usage: {
             prompt_tokens: (_h = objectResponse.usage.inputTokens) != null ? _h : 0,
             completion_tokens: (_i = objectResponse.usage.outputTokens) != null ? _i : 0,
-            total_tokens: (_j = objectResponse.usage.totalTokens) != null ? _j : 0
+            reasoning_tokens: (_j = objectResponse.usage.reasoningTokens) != null ? _j : 0,
+            cached_input_tokens: (_k = objectResponse.usage.cachedInputTokens) != null ? _k : 0,
+            total_tokens: (_l = objectResponse.usage.totalTokens) != null ? _l : 0
           }
         };
-        (_k = this.logger) == null ? void 0 : _k.call(this, {
+        (_m = this.logger) == null ? void 0 : _m.call(this, {
           category: "aisdk",
           message: "response",
           level: 1,
@@ -15871,12 +15920,14 @@ var AISdkClient = class extends LLMClient {
           }
         ],
         usage: {
-          prompt_tokens: (_l = textResponse.usage.inputTokens) != null ? _l : 0,
-          completion_tokens: (_m = textResponse.usage.outputTokens) != null ? _m : 0,
-          total_tokens: (_n = textResponse.usage.totalTokens) != null ? _n : 0
+          prompt_tokens: (_n = textResponse.usage.inputTokens) != null ? _n : 0,
+          completion_tokens: (_o = textResponse.usage.outputTokens) != null ? _o : 0,
+          reasoning_tokens: (_p = textResponse.usage.reasoningTokens) != null ? _p : 0,
+          cached_input_tokens: (_q = textResponse.usage.cachedInputTokens) != null ? _q : 0,
+          total_tokens: (_r = textResponse.usage.totalTokens) != null ? _r : 0
         }
       };
-      (_o = this.logger) == null ? void 0 : _o.call(this, {
+      (_s = this.logger) == null ? void 0 : _s.call(this, {
         category: "aisdk",
         message: "response",
         level: 2,
@@ -38723,7 +38774,7 @@ var AISdkClient2 = class extends LLMClient {
     return __async(this, arguments, function* ({
       options
     }) {
-      var _a, _b, _c, _d, _e, _f;
+      var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j;
       const formattedMessages = options.messages.map(
         (message) => {
           if (Array.isArray(message.content)) {
@@ -38784,7 +38835,9 @@ var AISdkClient2 = class extends LLMClient {
           usage: {
             prompt_tokens: (_a = response2.usage.inputTokens) != null ? _a : 0,
             completion_tokens: (_b = response2.usage.outputTokens) != null ? _b : 0,
-            total_tokens: (_c = response2.usage.totalTokens) != null ? _c : 0
+            reasoning_tokens: (_c = response2.usage.reasoningTokens) != null ? _c : 0,
+            cached_input_tokens: (_d = response2.usage.cachedInputTokens) != null ? _d : 0,
+            total_tokens: (_e = response2.usage.totalTokens) != null ? _e : 0
           }
         };
       }
@@ -38803,9 +38856,11 @@ var AISdkClient2 = class extends LLMClient {
       return {
         data: response.text,
         usage: {
-          prompt_tokens: (_d = response.usage.inputTokens) != null ? _d : 0,
-          completion_tokens: (_e = response.usage.outputTokens) != null ? _e : 0,
-          total_tokens: (_f = response.usage.totalTokens) != null ? _f : 0
+          prompt_tokens: (_f = response.usage.inputTokens) != null ? _f : 0,
+          completion_tokens: (_g = response.usage.outputTokens) != null ? _g : 0,
+          reasoning_tokens: (_h = response.usage.reasoningTokens) != null ? _h : 0,
+          cached_input_tokens: (_i = response.usage.cachedInputTokens) != null ? _i : 0,
+          total_tokens: (_j = response.usage.totalTokens) != null ? _j : 0
         }
       };
     });
@@ -39879,18 +39934,28 @@ var StagehandAPIClient = class {
       const metrics = {
         actPromptTokens: 0,
         actCompletionTokens: 0,
+        actReasoningTokens: 0,
+        actCachedInputTokens: 0,
         actInferenceTimeMs: 0,
         extractPromptTokens: 0,
         extractCompletionTokens: 0,
+        extractReasoningTokens: 0,
+        extractCachedInputTokens: 0,
         extractInferenceTimeMs: 0,
         observePromptTokens: 0,
         observeCompletionTokens: 0,
+        observeReasoningTokens: 0,
+        observeCachedInputTokens: 0,
         observeInferenceTimeMs: 0,
         agentPromptTokens: 0,
         agentCompletionTokens: 0,
+        agentReasoningTokens: 0,
+        agentCachedInputTokens: 0,
         agentInferenceTimeMs: 0,
         totalPromptTokens: 0,
         totalCompletionTokens: 0,
+        totalReasoningTokens: 0,
+        totalCachedInputTokens: 0,
         totalInferenceTimeMs: 0
       };
       const pages = apiData.pages || [];
@@ -39902,26 +39967,38 @@ var StagehandAPIClient = class {
           if (tokenUsage) {
             const inputTokens = tokenUsage.inputTokens || 0;
             const outputTokens = tokenUsage.outputTokens || 0;
+            const reasoningTokens = tokenUsage.reasoningTokens || 0;
+            const cachedInputTokens = tokenUsage.cachedInputTokens || 0;
             const timeMs = tokenUsage.timeMs || 0;
             if (method === "act") {
               metrics.actPromptTokens += inputTokens;
               metrics.actCompletionTokens += outputTokens;
+              metrics.actReasoningTokens += reasoningTokens;
+              metrics.actCachedInputTokens += cachedInputTokens;
               metrics.actInferenceTimeMs += timeMs;
             } else if (method === "extract") {
               metrics.extractPromptTokens += inputTokens;
               metrics.extractCompletionTokens += outputTokens;
+              metrics.extractReasoningTokens += reasoningTokens;
+              metrics.extractCachedInputTokens += cachedInputTokens;
               metrics.extractInferenceTimeMs += timeMs;
             } else if (method === "observe") {
               metrics.observePromptTokens += inputTokens;
               metrics.observeCompletionTokens += outputTokens;
+              metrics.observeReasoningTokens += reasoningTokens;
+              metrics.observeCachedInputTokens += cachedInputTokens;
               metrics.observeInferenceTimeMs += timeMs;
             } else if (method === "agent") {
               metrics.agentPromptTokens += inputTokens;
               metrics.agentCompletionTokens += outputTokens;
+              metrics.agentReasoningTokens += reasoningTokens;
+              metrics.agentCachedInputTokens += cachedInputTokens;
               metrics.agentInferenceTimeMs += timeMs;
             }
             metrics.totalPromptTokens += inputTokens;
             metrics.totalCompletionTokens += outputTokens;
+            metrics.totalReasoningTokens += reasoningTokens;
+            metrics.totalCachedInputTokens += cachedInputTokens;
             metrics.totalInferenceTimeMs += timeMs;
           }
         }
@@ -40090,18 +40167,28 @@ var _V3 = class _V3 {
     this.stagehandMetrics = {
       actPromptTokens: 0,
       actCompletionTokens: 0,
+      actReasoningTokens: 0,
+      actCachedInputTokens: 0,
       actInferenceTimeMs: 0,
       extractPromptTokens: 0,
       extractCompletionTokens: 0,
+      extractReasoningTokens: 0,
+      extractCachedInputTokens: 0,
       extractInferenceTimeMs: 0,
       observePromptTokens: 0,
       observeCompletionTokens: 0,
+      observeReasoningTokens: 0,
+      observeCachedInputTokens: 0,
       observeInferenceTimeMs: 0,
       agentPromptTokens: 0,
       agentCompletionTokens: 0,
+      agentReasoningTokens: 0,
+      agentCachedInputTokens: 0,
       agentInferenceTimeMs: 0,
       totalPromptTokens: 0,
       totalCompletionTokens: 0,
+      totalReasoningTokens: 0,
+      totalCachedInputTokens: 0,
       totalInferenceTimeMs: 0
     };
     var _a, _b, _c, _d, _e, _f, _g;
@@ -40285,34 +40372,50 @@ var _V3 = class _V3 {
       timestamp: (/* @__PURE__ */ new Date()).toISOString()
     });
   }
-  updateMetrics(functionName, promptTokens, completionTokens, inferenceTimeMs) {
+  updateMetrics(functionName, promptTokens, completionTokens, reasoningTokens, cachedInputTokens, inferenceTimeMs) {
     switch (functionName) {
       case "ACT" /* ACT */:
         this.stagehandMetrics.actPromptTokens += promptTokens;
         this.stagehandMetrics.actCompletionTokens += completionTokens;
+        this.stagehandMetrics.actReasoningTokens += reasoningTokens;
+        this.stagehandMetrics.actCachedInputTokens += cachedInputTokens;
         this.stagehandMetrics.actInferenceTimeMs += inferenceTimeMs;
         break;
       case "EXTRACT" /* EXTRACT */:
         this.stagehandMetrics.extractPromptTokens += promptTokens;
         this.stagehandMetrics.extractCompletionTokens += completionTokens;
+        this.stagehandMetrics.extractReasoningTokens += reasoningTokens;
+        this.stagehandMetrics.extractCachedInputTokens += cachedInputTokens;
         this.stagehandMetrics.extractInferenceTimeMs += inferenceTimeMs;
         break;
       case "OBSERVE" /* OBSERVE */:
         this.stagehandMetrics.observePromptTokens += promptTokens;
         this.stagehandMetrics.observeCompletionTokens += completionTokens;
+        this.stagehandMetrics.observeReasoningTokens += reasoningTokens;
+        this.stagehandMetrics.observeCachedInputTokens += cachedInputTokens;
         this.stagehandMetrics.observeInferenceTimeMs += inferenceTimeMs;
         break;
       case "AGENT" /* AGENT */:
         this.stagehandMetrics.agentPromptTokens += promptTokens;
         this.stagehandMetrics.agentCompletionTokens += completionTokens;
+        this.stagehandMetrics.agentReasoningTokens += reasoningTokens;
+        this.stagehandMetrics.agentCachedInputTokens += cachedInputTokens;
         this.stagehandMetrics.agentInferenceTimeMs += inferenceTimeMs;
         break;
     }
-    this.updateTotalMetrics(promptTokens, completionTokens, inferenceTimeMs);
+    this.updateTotalMetrics(
+      promptTokens,
+      completionTokens,
+      reasoningTokens,
+      cachedInputTokens,
+      inferenceTimeMs
+    );
   }
-  updateTotalMetrics(promptTokens, completionTokens, inferenceTimeMs) {
+  updateTotalMetrics(promptTokens, completionTokens, reasoningTokens, cachedInputTokens, inferenceTimeMs) {
     this.stagehandMetrics.totalPromptTokens += promptTokens;
     this.stagehandMetrics.totalCompletionTokens += completionTokens;
+    this.stagehandMetrics.totalReasoningTokens += reasoningTokens;
+    this.stagehandMetrics.totalCachedInputTokens += cachedInputTokens;
     this.stagehandMetrics.totalInferenceTimeMs += inferenceTimeMs;
   }
   _immediateShutdown(reason) {
@@ -40405,10 +40508,12 @@ var _V3 = class _V3 {
             (_a = this.opts.systemPrompt) != null ? _a : "",
             this.logInferenceToFile,
             (_b = this.opts.selfHeal) != null ? _b : true,
-            (functionName, promptTokens, completionTokens, inferenceTimeMs) => this.updateMetrics(
+            (functionName, promptTokens, completionTokens, reasoningTokens, cachedInputTokens, inferenceTimeMs) => this.updateMetrics(
               functionName,
               promptTokens,
               completionTokens,
+              reasoningTokens,
+              cachedInputTokens,
               inferenceTimeMs
             ),
             this.domSettleTimeoutMs
@@ -40421,10 +40526,12 @@ var _V3 = class _V3 {
             (_c = this.opts.systemPrompt) != null ? _c : "",
             this.logInferenceToFile,
             this.experimental,
-            (functionName, promptTokens, completionTokens, inferenceTimeMs) => this.updateMetrics(
+            (functionName, promptTokens, completionTokens, reasoningTokens, cachedInputTokens, inferenceTimeMs) => this.updateMetrics(
               functionName,
               promptTokens,
               completionTokens,
+              reasoningTokens,
+              cachedInputTokens,
               inferenceTimeMs
             )
           );
@@ -40436,10 +40543,12 @@ var _V3 = class _V3 {
             (_d = this.opts.systemPrompt) != null ? _d : "",
             this.logInferenceToFile,
             this.experimental,
-            (functionName, promptTokens, completionTokens, inferenceTimeMs) => this.updateMetrics(
+            (functionName, promptTokens, completionTokens, reasoningTokens, cachedInputTokens, inferenceTimeMs) => this.updateMetrics(
               functionName,
               promptTokens,
               completionTokens,
+              reasoningTokens,
+              cachedInputTokens,
               inferenceTimeMs
             )
           );

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@browserbasehq/stagehand",
-  "version": "3.0.2",
+  "version": "3.0.3-alpha-c76ade009ef81208accae6475ec4707d3906e566",
   "description": "An AI web browsing framework focused on simplicity and extensibility.",
   "main": "./dist/index.js",
   "module": "./dist/index.js",