@revenium/anthropic 1.1.3 → 1.1.4

package/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 All notable changes to this project will be documented in this file.
 
+## [1.1.4] - 2026-03-03
+
+### Fixed
+
+- Use canonical model name from API response for pricing resolution
+- Use resolved model in streaming error context
+
 ## [1.1.3] - 2026-02-19
 
 ### Changed

package/dist/cjs/wrapper.js

@@ -201,27 +201,29 @@ function reconstructResponseFromChunks(chunks, model) {
  * Handle streaming response by collecting chunks and extracting usage data
  */
 async function handleStreamingResponse(stream, context) {
-    const { requestId, model, metadata, requestTime, startTime, requestBody } = context;
-    // Create a new async generator that collects chunks and tracks usage
+    const { requestId, requestModel, metadata, requestTime, startTime, requestBody } = context;
     async function* trackingStream() {
         const chunks = [];
         let firstTokenTime;
+        let resolvedModel;
         try {
             for await (const chunk of stream) {
-                // Track first token time
                 if (!firstTokenTime && chunk.type === "content_block_delta") {
                     firstTokenTime = Date.now();
                 }
+                if (!resolvedModel && chunk.type === "message_start" && chunk.message?.model) {
+                    resolvedModel = chunk.message.model;
+                }
                 chunks.push(chunk);
                 yield chunk;
             }
-            // After stream completes, extract usage and track
             const endTime = Date.now();
             const responseTime = new Date();
             const duration = endTime - startTime;
             const timeToFirstToken = firstTokenTime
                 ? firstTokenTime - startTime
                 : undefined;
+            const model = resolvedModel ?? requestModel;
             logger.debug("Stream completed, extracting usage", {
                 requestId,
                 chunkCount: chunks.length,
@@ -233,7 +235,6 @@ async function handleStreamingResponse(stream, context) {
             if ((0, prompt_extraction_1.shouldCapturePrompts)(metadata)) {
                 reconstructedResponse = reconstructResponseFromChunks(chunks, model);
             }
-            // Create tracking data
             const trackingData = {
                 requestId,
                 model,
@@ -317,7 +318,7 @@ async function patchedCreateMethod(params, options) {
             // Create tracking data
             const trackingData = {
                 requestId,
-                model: params.model,
+                model: response.model ?? params.model,
                 inputTokens: usage.inputTokens,
                 outputTokens: usage.outputTokens,
                 cacheCreationTokens: usage.cacheCreationTokens,
@@ -336,7 +337,7 @@ async function patchedCreateMethod(params, options) {
             (0, tracking_1.trackUsageAsync)(trackingData);
             logger.debug("Anthropic request completed successfully", {
                 requestId,
-                model: params.model,
+                model: response.model ?? params.model,
                 inputTokens: usage.inputTokens,
                 outputTokens: usage.outputTokens,
                 duration,
@@ -346,7 +347,7 @@ async function patchedCreateMethod(params, options) {
         // Handle streaming response - need to collect chunks and extract usage
         return handleStreamingResponse(response, {
             requestId,
-            model: params.model,
+            requestModel: params.model,
             metadata,
             requestTime,
             startTime,
@@ -375,12 +376,12 @@ async function* patchedStreamMethod(params, options) {
     const responseTime = new Date();
     const chunks = [];
     let firstTokenTime;
+    let resolvedModel;
     logger.debug("Intercepted Anthropic messages.stream call", {
         requestId,
         model: params.model,
         hasMetadata: !!params.usageMetadata,
     });
-    // Validate parameters
     const validation = (0, validation_1.validateAnthropicMessageParams)(params);
     if (!validation.isValid) {
         logger.warn("Invalid Anthropic streaming parameters detected", {
@@ -389,22 +390,21 @@ async function* patchedStreamMethod(params, options) {
             warnings: validation.warnings,
         });
     }
-    // Extract and validate metadata
     const metadata = (0, validation_1.validateUsageMetadata)(params.usageMetadata || {});
-    // Remove usageMetadata from params before calling original method
     const { usageMetadata, ...cleanParams } = params;
     try {
-        // Call original stream method
         const originalStream = patchingContext.originalMethods?.stream;
         if (!originalStream) {
             throw new error_handling_1.StreamProcessingError("Original stream method not available");
         }
         const stream = originalStream.call(this, cleanParams, options);
         for await (const chunk of stream) {
-            // Track first token time
             if (!firstTokenTime && chunk.type === "content_block_delta") {
                 firstTokenTime = Date.now();
             }
+            if (!resolvedModel && chunk.type === "message_start" && chunk.message?.model) {
+                resolvedModel = chunk.message.model;
+            }
             chunks.push(chunk);
             yield chunk;
         }
@@ -413,18 +413,16 @@ async function* patchedStreamMethod(params, options) {
         const timeToFirstToken = firstTokenTime
             ? firstTokenTime - startTime
             : undefined;
-        // Extract usage information from all chunks
+        const model = resolvedModel ?? params.model;
         const usage = (0, tracking_1.extractUsageFromStream)(chunks);
-        // Detect vision content
         const hasVisionContent = (0, trace_fields_1.detectVisionContent)(params);
         let reconstructedResponse = undefined;
         if ((0, prompt_extraction_1.shouldCapturePrompts)(metadata)) {
-            reconstructedResponse = reconstructResponseFromChunks(chunks, params.model);
+            reconstructedResponse = reconstructResponseFromChunks(chunks, model);
         }
-        // Create tracking data
         const trackingData = {
             requestId,
-            model: params.model,
+            model,
             inputTokens: usage.inputTokens,
             outputTokens: usage.outputTokens,
             cacheCreationTokens: usage.cacheCreationTokens,
@@ -440,11 +438,10 @@ async function* patchedStreamMethod(params, options) {
             requestBody: params,
             response: reconstructedResponse,
         };
-        // Track usage asynchronously
         (0, tracking_1.trackUsageAsync)(trackingData);
         logger.debug("Anthropic streaming request completed successfully", {
             requestId,
-            model: params.model,
+            model,
             inputTokens: usage.inputTokens,
             outputTokens: usage.outputTokens,
             duration,
@@ -457,7 +454,7 @@ async function* patchedStreamMethod(params, options) {
         const duration = endTime - startTime;
         const errorContext = (0, error_handling_1.createErrorContext)()
             .withRequestId(requestId)
-            .withModel(params.model)
+            .withModel(resolvedModel ?? params.model)
             .withDuration(duration)
             .with("isStreaming", true)
             .with("chunkCount", chunks.length)

package/dist/esm/wrapper.js

@@ -193,27 +193,29 @@ function reconstructResponseFromChunks(chunks, model) {
  * Handle streaming response by collecting chunks and extracting usage data
  */
 async function handleStreamingResponse(stream, context) {
-    const { requestId, model, metadata, requestTime, startTime, requestBody } = context;
-    // Create a new async generator that collects chunks and tracks usage
+    const { requestId, requestModel, metadata, requestTime, startTime, requestBody } = context;
     async function* trackingStream() {
         const chunks = [];
         let firstTokenTime;
+        let resolvedModel;
         try {
             for await (const chunk of stream) {
-                // Track first token time
                 if (!firstTokenTime && chunk.type === "content_block_delta") {
                     firstTokenTime = Date.now();
                 }
+                if (!resolvedModel && chunk.type === "message_start" && chunk.message?.model) {
+                    resolvedModel = chunk.message.model;
+                }
                 chunks.push(chunk);
                 yield chunk;
             }
-            // After stream completes, extract usage and track
             const endTime = Date.now();
             const responseTime = new Date();
             const duration = endTime - startTime;
             const timeToFirstToken = firstTokenTime
                 ? firstTokenTime - startTime
                 : undefined;
+            const model = resolvedModel ?? requestModel;
             logger.debug("Stream completed, extracting usage", {
                 requestId,
                 chunkCount: chunks.length,
@@ -225,7 +227,6 @@ async function handleStreamingResponse(stream, context) {
             if (shouldCapturePrompts(metadata)) {
                 reconstructedResponse = reconstructResponseFromChunks(chunks, model);
             }
-            // Create tracking data
             const trackingData = {
                 requestId,
                 model,
@@ -309,7 +310,7 @@ async function patchedCreateMethod(params, options) {
             // Create tracking data
             const trackingData = {
                 requestId,
-                model: params.model,
+                model: response.model ?? params.model,
                 inputTokens: usage.inputTokens,
                 outputTokens: usage.outputTokens,
                 cacheCreationTokens: usage.cacheCreationTokens,
@@ -328,7 +329,7 @@ async function patchedCreateMethod(params, options) {
             trackUsageAsync(trackingData);
             logger.debug("Anthropic request completed successfully", {
                 requestId,
-                model: params.model,
+                model: response.model ?? params.model,
                 inputTokens: usage.inputTokens,
                 outputTokens: usage.outputTokens,
                 duration,
@@ -338,7 +339,7 @@ async function patchedCreateMethod(params, options) {
         // Handle streaming response - need to collect chunks and extract usage
         return handleStreamingResponse(response, {
             requestId,
-            model: params.model,
+            requestModel: params.model,
             metadata,
             requestTime,
             startTime,
@@ -367,12 +368,12 @@ async function* patchedStreamMethod(params, options) {
     const responseTime = new Date();
     const chunks = [];
     let firstTokenTime;
+    let resolvedModel;
     logger.debug("Intercepted Anthropic messages.stream call", {
         requestId,
         model: params.model,
         hasMetadata: !!params.usageMetadata,
     });
-    // Validate parameters
     const validation = validateAnthropicMessageParams(params);
     if (!validation.isValid) {
         logger.warn("Invalid Anthropic streaming parameters detected", {
@@ -381,22 +382,21 @@ async function* patchedStreamMethod(params, options) {
             warnings: validation.warnings,
         });
     }
-    // Extract and validate metadata
     const metadata = validateUsageMetadata(params.usageMetadata || {});
-    // Remove usageMetadata from params before calling original method
     const { usageMetadata, ...cleanParams } = params;
     try {
-        // Call original stream method
         const originalStream = patchingContext.originalMethods?.stream;
         if (!originalStream) {
             throw new StreamProcessingError("Original stream method not available");
         }
         const stream = originalStream.call(this, cleanParams, options);
         for await (const chunk of stream) {
-            // Track first token time
             if (!firstTokenTime && chunk.type === "content_block_delta") {
                 firstTokenTime = Date.now();
             }
+            if (!resolvedModel && chunk.type === "message_start" && chunk.message?.model) {
+                resolvedModel = chunk.message.model;
+            }
             chunks.push(chunk);
             yield chunk;
         }
@@ -405,18 +405,16 @@ async function* patchedStreamMethod(params, options) {
         const timeToFirstToken = firstTokenTime
             ? firstTokenTime - startTime
             : undefined;
-        // Extract usage information from all chunks
+        const model = resolvedModel ?? params.model;
         const usage = extractUsageFromStream(chunks);
-        // Detect vision content
         const hasVisionContent = detectVisionContent(params);
         let reconstructedResponse = undefined;
         if (shouldCapturePrompts(metadata)) {
-            reconstructedResponse = reconstructResponseFromChunks(chunks, params.model);
+            reconstructedResponse = reconstructResponseFromChunks(chunks, model);
         }
-        // Create tracking data
         const trackingData = {
             requestId,
-            model: params.model,
+            model,
             inputTokens: usage.inputTokens,
             outputTokens: usage.outputTokens,
             cacheCreationTokens: usage.cacheCreationTokens,
@@ -432,11 +430,10 @@ async function* patchedStreamMethod(params, options) {
             requestBody: params,
             response: reconstructedResponse,
         };
-        // Track usage asynchronously
         trackUsageAsync(trackingData);
         logger.debug("Anthropic streaming request completed successfully", {
             requestId,
-            model: params.model,
+            model,
             inputTokens: usage.inputTokens,
             outputTokens: usage.outputTokens,
             duration,
@@ -449,7 +446,7 @@ async function* patchedStreamMethod(params, options) {
         const duration = endTime - startTime;
         const errorContext = createErrorContext()
             .withRequestId(requestId)
-            .withModel(params.model)
+            .withModel(resolvedModel ?? params.model)
             .withDuration(duration)
             .with("isStreaming", true)
             .with("chunkCount", chunks.length)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@revenium/anthropic",
-  "version": "1.1.3",
+  "version": "1.1.4",
   "description": "Transparent TypeScript middleware for automatic Revenium usage tracking with Anthropic Claude AI",
   "main": "dist/cjs/index.js",
   "module": "dist/esm/index.js",