diff --git a/server/gemma4_test.go b/server/gemma4_test.go
index 9ad3b0ce4..945fd4be9 100644
--- a/server/gemma4_test.go
+++ b/server/gemma4_test.go
@@ -80,6 +80,15 @@ func TestResolveGemma4Renderer(t *testing.T) {
 			},
 			want: gemma4RendererLarge,
 		},
+		{
+			name: "legacy 11.9B model type with size-free name resolves large",
+			model: &Model{
+				Name:      "my-gemma",
+				ShortName: "my-gemma",
+				Config:    testConfigWithRendererAndType(gemma4RendererLegacy, "11.9B"),
+			},
+			want: gemma4RendererLarge,
+		},
 		{
 			name: "legacy unknown defaults small",
 			model: &Model{
diff --git a/server/renderer_resolution.go b/server/renderer_resolution.go
index 53f87355a..3f0b4f07c 100644
--- a/server/renderer_resolution.go
+++ b/server/renderer_resolution.go
@@ -14,8 +14,9 @@ const (
 
 	// Gemma 4 small templates cover the e2b/e4b family, while 12b/26b/31b use
 	// the large template. Default to the small prompt unless the model is
-	// clearly in the large range.
-	gemma4LargeMinParameterCount = 12_000_000_000
+	// clearly in the large range. The 12b model reports 11.9B parameters, so
+	// the threshold sits just below that and above e4b.
+	gemma4LargeMinParameterCount = 11_500_000_000
 )
 
 func resolveRendererName(m *Model) string {