diff --git a/docs/_static/llama-stack-spec.html b/docs/_static/llama-stack-spec.html
index 2a294ea11..15b06257f 100644
--- a/docs/_static/llama-stack-spec.html
+++ b/docs/_static/llama-stack-spec.html
@@ -8653,7 +8653,7 @@
             "EvaluationResponse": {
                 "type": "object",
                 "properties": {
-                    "result_data": {
+                    "result_rows": {
                         "type": "array",
                         "items": {
                             "type": "object",
@@ -8680,9 +8680,9 @@
                                 ]
                             }
                         },
-                        "description": "The result data containing generations and grades in each row."
+                        "description": "The result data containing inputs, generations and grades in each row."
                     },
-                    "metrics": {
+                    "grades": {
                         "type": "object",
                         "additionalProperties": {
                             "oneOf": [
@@ -8706,13 +8706,13 @@
                                 }
                             ]
                         },
-                        "description": "Map of metric name to aggregated value."
+                        "description": "Map of grader id to aggregated value."
                     }
                 },
                 "additionalProperties": false,
                 "required": [
-                    "result_data",
-                    "metrics"
+                    "result_rows",
+                    "grades"
                 ],
                 "title": "EvaluationResponse",
                 "description": "A response to an inline evaluation."
diff --git a/docs/_static/llama-stack-spec.yaml b/docs/_static/llama-stack-spec.yaml
index 7508acd66..0f83dd3d7 100644
--- a/docs/_static/llama-stack-spec.yaml
+++ b/docs/_static/llama-stack-spec.yaml
@@ -6018,7 +6018,7 @@ components:
     EvaluationResponse:
       type: object
       properties:
-        result_data:
+        result_rows:
           type: array
           items:
             type: object
@@ -6031,8 +6031,8 @@ components:
                 - type: array
                 - type: object
           description: >-
-            The result data containing generations and grades in each row.
-        metrics:
+            The result data containing inputs, generations and grades in each row.
+        grades:
           type: object
           additionalProperties:
             oneOf:
@@ -6042,11 +6042,11 @@ components:
               - type: string
               - type: array
               - type: object
-          description: Map of metric name to aggregated value.
+          description: Map of grader id to aggregated value.
       additionalProperties: false
       required:
-        - result_data
-        - metrics
+        - result_rows
+        - grades
       title: EvaluationResponse
       description: A response to an inline evaluation.
     HealthInfo:
diff --git a/llama_stack/apis/evaluation/evaluation.py b/llama_stack/apis/evaluation/evaluation.py
index 8d6fdd201..bde27e0be 100644
--- a/llama_stack/apis/evaluation/evaluation.py
+++ b/llama_stack/apis/evaluation/evaluation.py
@@ -81,12 +81,12 @@ class EvaluationResponse(BaseModel):
     """
     A response to an inline evaluation.
 
-    :param result_data: The result data containing generations and grades in each row.
-    :param metrics: Map of metric name to aggregated value.
+    :param result_rows: The result data containing inputs, generations and grades in each row.
+    :param grades: Map of grader id to aggregated value.
     """
 
-    result_data: List[Dict[str, Any]]
-    metrics: Dict[str, Any]
+    result_rows: List[Dict[str, Any]]
+    grades: Dict[str, Any]
 
 
 class Evaluation(Protocol):