diff --git a/go/adk/pkg/embedding/embedding.go b/go/adk/pkg/embedding/embedding.go
index 371a1de9e..44b5ab000 100644
--- a/go/adk/pkg/embedding/embedding.go
+++ b/go/adk/pkg/embedding/embedding.go
@@ -115,9 +115,8 @@ func generateEmbeddings(ctx context.Context, client openai.Client, cfg *adk.Embe
 	log := logr.FromContextOrDiscard(ctx)
 
 	resp, err := client.Embeddings.New(ctx, openai.EmbeddingNewParams{
-		Model:      openai.EmbeddingModel(cfg.Model),
-		Input:      openai.EmbeddingNewParamsInputUnion{OfArrayOfStrings: texts},
-		Dimensions: openai.Int(int64(TargetDimension)),
+		Model: openai.EmbeddingModel(cfg.Model),
+		Input: openai.EmbeddingNewParamsInputUnion{OfArrayOfStrings: texts},
 	}, embeddingPassthroughOpts(ctx, cfg, isAzureFamily)...)
 	if err != nil {
 		return nil, fmt.Errorf("%s embeddings request failed: %w", provider, err)
diff --git a/go/adk/pkg/embedding/embedding_test.go b/go/adk/pkg/embedding/embedding_test.go
index 63ce20c38..220d36ad8 100644
--- a/go/adk/pkg/embedding/embedding_test.go
+++ b/go/adk/pkg/embedding/embedding_test.go
@@ -29,9 +29,8 @@ func TestOpenAIProvider_UsesAPIKeyNotKagentToken(t *testing.T) {
 			t.Fatalf("read body: %v", err)
 		}
 		var req struct {
-			Model      string   `json:"model"`
-			Input      []string `json:"input"`
-			Dimensions int      `json:"dimensions"`
+			Model string   `json:"model"`
+			Input []string `json:"input"`
 		}
 		if err := json.Unmarshal(body, &req); err != nil {
 			t.Fatalf("unmarshal request: %v", err)
@@ -39,8 +38,18 @@ func TestOpenAIProvider_UsesAPIKeyNotKagentToken(t *testing.T) {
 		if req.Model != "text-embedding-3-small" {
 			t.Errorf("model = %q, want text-embedding-3-small", req.Model)
 		}
-		if req.Dimensions != TargetDimension {
-			t.Errorf("dimensions = %d, want %d", req.Dimensions, TargetDimension)
+		// dimensions must not be sent at all: providers that don't offer
+		// exactly TargetDimension (e.g. Bedrock Titan behind an
+		// OpenAI-compatible gateway) would reject the request outright.
+		// processEmbeddings truncates whatever the provider returns instead.
+		// Checked against the raw body, not a decoded field, since decoding
+		// into an int can't tell an absent field from an explicit 0.
+		var rawFields map[string]json.RawMessage
+		if err := json.Unmarshal(body, &rawFields); err != nil {
+			t.Fatalf("unmarshal request as map: %v", err)
+		}
+		if _, present := rawFields["dimensions"]; present {
+			t.Errorf("request body has a dimensions field, want it omitted: %s", body)
 		}
 		if len(req.Input) != 1 || req.Input[0] != "hello" {
 			t.Errorf("input = %v, want [hello]", req.Input)
diff --git a/python/packages/kagent-adk/src/kagent/adk/models/_embedding.py b/python/packages/kagent-adk/src/kagent/adk/models/_embedding.py
index deccc2978..e7481d0bc 100644
--- a/python/packages/kagent-adk/src/kagent/adk/models/_embedding.py
+++ b/python/packages/kagent-adk/src/kagent/adk/models/_embedding.py
@@ -174,7 +174,6 @@ async def _embed_openai(self, texts: List[str]) -> List[List[float]]:
         response = await client.embeddings.create(
             model=self.config.model,
             input=texts,
-            dimensions=self.TARGET_DIMENSION,
         )
         return [item.embedding for item in response.data]
 
diff --git a/python/packages/kagent-adk/tests/unittests/test_embedding.py b/python/packages/kagent-adk/tests/unittests/test_embedding.py
index 0144f6fdf..cc4363a01 100644
--- a/python/packages/kagent-adk/tests/unittests/test_embedding.py
+++ b/python/packages/kagent-adk/tests/unittests/test_embedding.py
@@ -53,6 +53,24 @@ async def test_generate_single_text(self):
             result = await client.generate("hello world")
         assert result == vec
 
+    @pytest.mark.asyncio
+    async def test_generate_single_text_omits_dimensions(self):
+        # Providers that don't offer exactly 768 dimensions (e.g. Bedrock Titan
+        # behind an OpenAI-compatible gateway) reject the request outright when
+        # dimensions is set. _process_embeddings truncates whatever comes back.
+        client = make_client(provider="openai", model="text-embedding-3-small")
+        vec = [0.1] * 768
+        mock_response = make_openai_embedding_response([vec])
+        with mock.patch("openai.AsyncOpenAI") as mock_cls:
+            instance = mock.AsyncMock()
+            instance.embeddings.create = mock.AsyncMock(return_value=mock_response)
+            mock_cls.return_value = instance
+            await client.generate("hello world")
+        instance.embeddings.create.assert_called_once_with(
+            model="text-embedding-3-small",
+            input=["hello world"],
+        )
+
     @pytest.mark.asyncio
     async def test_generate_batch_texts(self):
         client = make_client(provider="openai", model="text-embedding-3-small")