fix(langchain): fix structured output on opus (#10010)

Author: christian-bromann

.changeset/thirty-readers-sniff.md

@@ -0,0 +1,5 @@
+---
+"langchain": patch
+---
+
+fix(langchain): fix structured output on opus

.github/workflows/benchmark-tests.yml

@@ -4,13 +4,13 @@
 name: Benchmark Tests (LangChain)
 
 on:
-  push:
-    branches: ["main"]
-  pull_request:
-    # Only run when LangChain, LangChain Core, or LangChain Classic code changes.
-    paths:
-      - "libs/langchain/**"
-      - "libs/langchain-core/**"
+  # push:
+  #   branches: ["main"]
+  # pull_request:
+  #   # Only run when LangChain, LangChain Core, or LangChain Classic code changes.
+  #   paths:
+  #     - "libs/langchain/**"
+  #     - "libs/langchain-core/**"
   workflow_dispatch: # Allows triggering the workflow manually in GitHub UI
 
 # If another push to the same PR or branch happens while this workflow is still running,

.vscode/settings.json

@@ -36,5 +36,6 @@
   },
   "[typescript]": {
     "editor.defaultFormatter": "esbenp.prettier-vscode"
-  }
+  },
+  "deno.enable": false
 }

libs/langchain/src/agents/nodes/AgentNode.ts

@@ -909,17 +909,28 @@ export class AgentNode<
       };
 
       Object.assign(options, {
+        /**
+         * OpenAI-style options
+         * Used by ChatOpenAI, ChatXAI, and other OpenAI-compatible providers.
+         */
         response_format: {
           type: "json_schema",
           json_schema: jsonSchemaParams,
         },
-        output_format: {
-          type: "json_schema",
-          schema: structuredResponseFormat.strategy.schema,
-        },
-        headers: {
-          "anthropic-beta": "structured-outputs-2025-11-13",
+
+        /**
+         * Anthropic-style options
+         */
+        outputConfig: {
+          format: {
+            type: "json_schema",
+            schema: structuredResponseFormat.strategy.schema,
+          },
         },
+
+        /**
+         * for LangSmith structured output tracing
+         */
         ls_structured_output_format: {
           kwargs: { method: "json_schema" },
           schema: structuredResponseFormat.strategy.schema,

libs/langchain/src/agents/tests/responses.int.test.ts

@@ -264,6 +264,59 @@ describe("structured output handling", () => {
       expect(result.messages.at(-1)?.content).toContain('{"answer": "Paris"}');
       expect((result.messages.at(-1) as AIMessage).tool_calls?.length).toBe(0);
     });
+
+    it("should return structured output with Claude Opus 4.6 end-to-end", async () => {
+      const model = new ChatAnthropic({
+        model: "claude-opus-4-6",
+      });
+
+      const { tool: weatherTool, mock: weatherMock } = makeTool(
+        ({ city }: { city: string }) =>
+          `The weather in ${city} is 22°C and partly cloudy.`,
+        {
+          name: "get_weather",
+          description: "Get the current weather for a city",
+          schema: z.object({
+            city: z.string().describe("The city name"),
+          }),
+        }
+      );
+
+      const ResponseSchema = z.object({
+        city: z.string().describe("The city that was queried"),
+        temperature: z.number().describe("The temperature"),
+        summary: z.string().describe("A brief weather summary"),
+      });
+
+      const agent = createAgent({
+        model,
+        tools: [weatherTool],
+        systemPrompt:
+          "You are a weather assistant. Use the get_weather tool to answer questions about weather. Return exact values from the tool.",
+        responseFormat: providerStrategy(ResponseSchema),
+      });
+
+      const result = await agent.invoke({
+        messages: [new HumanMessage("What is the weather in Berlin?")],
+      });
+
+      // Verify tool was called
+      expect(weatherMock).toHaveBeenCalledTimes(1);
+
+      // Verify structured response is present and correctly parsed
+      expect(result.structuredResponse).toBeDefined();
+      expect(result.structuredResponse.temperature).toBe(22);
+      expect(result.structuredResponse.city).toBe("Berlin");
+      expect(typeof result.structuredResponse.summary).toBe("string");
+
+      // Verify no extract- tool calls (providerStrategy uses native structured output)
+      const hasExtractToolCalls = result.messages.some(
+        (msg) =>
+          AIMessage.isInstance(msg) &&
+          msg.tool_calls?.some((tc) => tc.name.startsWith("extract-"))
+      );
+      expect(hasExtractToolCalls).toBe(false);
+    });
   });
 });