fix(gateway): disable thinking for qwen3.6-35b-a3b json mode

Re-enables jsonOutput on the model and disables thinking only when
JSON response_format is requested, since combining both upstream
returns empty content ~40% of the time.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: steebchen

packages/actions/src/prepare-request-body.ts

@@ -1031,6 +1031,17 @@ export async function prepareRequestBody(
 		}
 	}
 
+	// qwen3.6-35b-a3b returns empty content ~40% of the time when json_object
+	// is combined with thinking; force thinking off when JSON is requested.
+	if (
+		usedProvider === "alibaba" &&
+		usedModel === "qwen3.6-35b-a3b" &&
+		(response_format?.type === "json_object" ||
+			response_format?.type === "json_schema")
+	) {
+		requestBody.enable_thinking = false;
+	}
+
 	if (forcesToolUse && usedProvider === "moonshot") {
 		const providerMapping = modelDef?.providers.find(
 			(p) => p.modelName === usedModel && p.providerId === usedProvider,

packages/models/src/models/alibaba.ts

@@ -2357,9 +2357,7 @@ export const alibabaModels = [
 				tools: true,
 				webSearch: true,
 				webSearchPrice: 0.01,
-				// json_object mode in thinking mode returns empty content ~40% of the time;
-				// disabled until Alibaba fixes the upstream behavior
-				jsonOutput: false,
+				jsonOutput: true,
 				// Qwen thinking models reject tool_choice "required" or object
 				supportedParameters: [
 					"temperature",
@@ -2369,6 +2367,7 @@ export const alibabaModels = [
 					"presence_penalty",
 					"stop",
 					"stream",
+					"response_format",
 					"tools",
 				],
 			},