diff --git a/.changeset/calm-squids-sparkle.md b/.changeset/calm-squids-sparkle.md
new file mode 100644
index 000000000000..51ea1832f074
--- /dev/null
+++ b/.changeset/calm-squids-sparkle.md
@@ -0,0 +1,5 @@
+---
+'ai': patch
+---
+
+fix(ai): doStream should reflect transformed values
diff --git a/packages/ai/src/generate-text/__snapshots__/stream-text.test.ts.snap b/packages/ai/src/generate-text/__snapshots__/stream-text.test.ts.snap
index 7960b8905280..6e300109f9af 100644
--- a/packages/ai/src/generate-text/__snapshots__/stream-text.test.ts.snap
+++ b/packages/ai/src/generate-text/__snapshots__/stream-text.test.ts.snap
@@ -240,8 +240,8 @@ exports[`streamText > options.transform > with base transformation > telemetry s
       "ai.response.model": "mock-model-id",
       "ai.response.msToFinish": 500,
       "ai.response.msToFirstChunk": 100,
-      "ai.response.providerMetadata": "{"testProvider":{"testKey":"testValue"}}",
-      "ai.response.text": "Hello, world!",
+      "ai.response.providerMetadata": "{"testProvider":{"testKey":"TEST VALUE"}}",
+      "ai.response.text": "HELLO, WORLD!",
       "ai.response.timestamp": "1970-01-01T00:00:00.000Z",
       "ai.response.toolCalls": "[{"type":"tool-call","toolCallId":"call-1","toolName":"tool1","input":{"value":"VALUE"}}]",
       "ai.settings.maxRetries": 2,
diff --git a/packages/ai/src/generate-text/stream-text.ts b/packages/ai/src/generate-text/stream-text.ts
index 74ff36841eb8..eca3f0389090 100644
--- a/packages/ai/src/generate-text/stream-text.ts
+++ b/packages/ai/src/generate-text/stream-text.ts
@@ -1946,29 +1946,13 @@ class DefaultStreamTextResult<TOOLS extends ToolSet, OUTPUT extends Output>
                         ? JSON.stringify(stepToolCalls)
                         : undefined;
 
-                    // record telemetry information first to ensure best effort timing
+                    // record telemetry attributes that don't depend on transforms:
                     try {
                       doStreamSpan.setAttributes(
                         await selectTelemetryAttributes({
                           telemetry,
                           attributes: {
                             'ai.response.finishReason': stepFinishReason,
-                            'ai.response.text': {
-                              output: () => activeText,
-                            },
-                            'ai.response.reasoning': {
-                              output: () => {
-                                const reasoningParts = recordedContent.filter(
-                                  (
-                                    c,
-                                  ): c is { type: 'reasoning'; text: string } =>
-                                    c.type === 'reasoning',
-                                );
-                                return reasoningParts.length > 0
-                                  ? reasoningParts.map(r => r.text).join('\n')
-                                  : undefined;
-                              },
-                            },
                             'ai.response.toolCalls': {
                               output: () => stepToolCallsJson,
                             },
@@ -1976,8 +1960,6 @@ class DefaultStreamTextResult<TOOLS extends ToolSet, OUTPUT extends Output>
                             'ai.response.model': stepResponse.modelId,
                             'ai.response.timestamp':
                               stepResponse.timestamp.toISOString(),
-                            'ai.response.providerMetadata':
-                              JSON.stringify(stepProviderMetadata),
 
                             'ai.usage.inputTokens': stepUsage.inputTokens,
                             'ai.usage.outputTokens': stepUsage.outputTokens,
@@ -2001,9 +1983,6 @@ class DefaultStreamTextResult<TOOLS extends ToolSet, OUTPUT extends Output>
                       );
                     } catch (error) {
                       // ignore error setting telemetry attributes
-                    } finally {
-                      // finish doStreamSpan before other operations for correct timing:
-                      doStreamSpan.end();
                     }
 
                     controller.enqueue({
@@ -2027,6 +2006,33 @@ class DefaultStreamTextResult<TOOLS extends ToolSet, OUTPUT extends Output>
                     // to ensure that the recorded steps are complete:
                     await stepFinish.promise;
 
+                    // set transform-dependent attributes after the step has been
+                    // fully processed (post-transform) by the event processor:
+                    const processedStep =
+                      recordedSteps[recordedSteps.length - 1];
+                    try {
+                      doStreamSpan.setAttributes(
+                        await selectTelemetryAttributes({
+                          telemetry,
+                          attributes: {
+                            'ai.response.text': {
+                              output: () => processedStep.text,
+                            },
+                            'ai.response.reasoning': {
+                              output: () => processedStep.reasoningText,
+                            },
+                            'ai.response.providerMetadata': JSON.stringify(
+                              processedStep.providerMetadata,
+                            ),
+                          },
+                        }),
+                      );
+                    } catch (error) {
+                      // ignore error setting telemetry attributes
+                    } finally {
+                      doStreamSpan.end();
+                    }
+
                     const clientToolCalls = stepToolCalls.filter(
                       toolCall => toolCall.providerExecuted !== true,
                     );