@langchain/openai 0.5.15 → 0.5.17

package/dist/chat_models.js

@@ -1,4 +1,3 @@
-import * as z4 from "zod/v4/core";
 import { OpenAI as OpenAIClient } from "openai";
 import { AIMessage, AIMessageChunk, ChatMessage, ChatMessageChunk, FunctionMessageChunk, HumanMessageChunk, SystemMessageChunk, ToolMessageChunk, isAIMessage, parseBase64DataUrl, parseMimeType, convertToProviderContentBlock, isDataContentBlock, } from "@langchain/core/messages";
 import { ChatGenerationChunk, } from "@langchain/core/outputs";
@@ -8,13 +7,33 @@ import { isOpenAITool, } from "@langchain/core/language_models/base";
 import { RunnableLambda, RunnablePassthrough, RunnableSequence, } from "@langchain/core/runnables";
 import { JsonOutputParser, StructuredOutputParser, } from "@langchain/core/output_parsers";
 import { JsonOutputKeyToolsParser, convertLangChainToolCallToOpenAI, makeInvalidToolCall, parseToolCall, } from "@langchain/core/output_parsers/openai_tools";
-import { zodResponseFormat } from "openai/helpers/zod";
-import { getSchemaDescription, isInteropZodSchema, isZodSchemaV3, isZodSchemaV4, } from "@langchain/core/utils/types";
+import { getSchemaDescription, isInteropZodSchema, } from "@langchain/core/utils/types";
 import { toJsonSchema, } from "@langchain/core/utils/json_schema";
 import { getEndpoint } from "./utils/azure.js";
-import { formatToOpenAIToolChoice, wrapOpenAIClientError, } from "./utils/openai.js";
+import { formatToOpenAIToolChoice, interopZodResponseFormat, wrapOpenAIClientError, } from "./utils/openai.js";
 import { formatFunctionDefinitions, } from "./utils/openai-format-fndef.js";
 import { _convertToOpenAITool } from "./utils/tools.js";
+const _FUNCTION_CALL_IDS_MAP_KEY = "__openai_function_call_ids__";
+function isBuiltInTool(tool) {
+    return "type" in tool && tool.type !== "function";
+}
+function isBuiltInToolChoice(tool_choice) {
+    return (tool_choice != null &&
+        typeof tool_choice === "object" &&
+        "type" in tool_choice &&
+        tool_choice.type !== "function");
+}
+function isReasoningModel(model) {
+    return model && /^o\d/.test(model);
+}
+function isStructuredOutputMethodParams(x
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+) {
+    return (x !== undefined &&
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        typeof x.schema ===
+            "object");
+}
 function extractGenericMessageCustomRole(message) {
     if (message.role !== "system" &&
         message.role !== "developer" &&
@@ -175,6 +194,7 @@ const completionsApiContentBlockConverter = {
     },
 };
 // Used in LangSmith, export is important here
+// TODO: put this conversion elsewhere
 export function _convertMessagesToOpenAIParams(messages, model) {
     // TODO: Function messages do not support array content, fix cast
     return messages.flatMap((message) => {
@@ -223,1114 +243,112 @@ export function _convertMessagesToOpenAIParams(messages, model) {
                     id: message.additional_kwargs.audio.id,
                 },
             };
-            return [completionParam, audioMessage];
-        }
-        return completionParam;
-    });
-}
-const _FUNCTION_CALL_IDS_MAP_KEY = "__openai_function_call_ids__";
-function _convertReasoningSummaryToOpenAIResponsesParams(reasoning) {
-    // combine summary parts that have the the same index and then remove the indexes
-    const summary = (reasoning.summary.length > 1
-        ? reasoning.summary.reduce((acc, curr) => {
-            const last = acc.at(-1);
-            if (last.index === curr.index) {
-                last.text += curr.text;
-            }
-            else {
-                acc.push(curr);
-            }
-            return acc;
-        }, [{ ...reasoning.summary[0] }])
-        : reasoning.summary).map((s) => Object.fromEntries(Object.entries(s).filter(([k]) => k !== "index")));
-    return {
-        ...reasoning,
-        summary,
-    };
-}
-function _convertMessagesToOpenAIResponsesParams(messages, model, zdrEnabled) {
-    return messages.flatMap((lcMsg) => {
-        const additional_kwargs = lcMsg.additional_kwargs;
-        let role = messageToOpenAIRole(lcMsg);
-        if (role === "system" && isReasoningModel(model))
-            role = "developer";
-        if (role === "function") {
-            throw new Error("Function messages are not supported in Responses API");
-        }
-        if (role === "tool") {
-            const toolMessage = lcMsg;
-            // Handle computer call output
-            if (additional_kwargs?.type === "computer_call_output") {
-                const output = (() => {
-                    if (typeof toolMessage.content === "string") {
-                        return {
-                            type: "computer_screenshot",
-                            image_url: toolMessage.content,
-                        };
-                    }
-                    if (Array.isArray(toolMessage.content)) {
-                        const oaiScreenshot = toolMessage.content.find((i) => i.type === "computer_screenshot");
-                        if (oaiScreenshot)
-                            return oaiScreenshot;
-                        const lcImage = toolMessage.content.find((i) => i.type === "image_url");
-                        if (lcImage) {
-                            return {
-                                type: "computer_screenshot",
-                                image_url: typeof lcImage.image_url === "string"
-                                    ? lcImage.image_url
-                                    : lcImage.image_url.url,
-                            };
-                        }
-                    }
-                    throw new Error("Invalid computer call output");
-                })();
-                return {
-                    type: "computer_call_output",
-                    output,
-                    call_id: toolMessage.tool_call_id,
-                };
-            }
-            return {
-                type: "function_call_output",
-                call_id: toolMessage.tool_call_id,
-                id: toolMessage.id?.startsWith("fc_") ? toolMessage.id : undefined,
-                output: typeof toolMessage.content !== "string"
-                    ? JSON.stringify(toolMessage.content)
-                    : toolMessage.content,
-            };
-        }
-        if (role === "assistant") {
-            // if we have the original response items, just reuse them
-            if (!zdrEnabled &&
-                lcMsg.response_metadata.output != null &&
-                Array.isArray(lcMsg.response_metadata.output) &&
-                lcMsg.response_metadata.output.length > 0 &&
-                lcMsg.response_metadata.output.every((item) => "type" in item)) {
-                return lcMsg.response_metadata.output;
-            }
-            // otherwise, try to reconstruct the response from what we have
-            const input = [];
-            // reasoning items
-            if (additional_kwargs?.reasoning && !zdrEnabled) {
-                const reasoningItem = _convertReasoningSummaryToOpenAIResponsesParams(additional_kwargs.reasoning);
-                input.push(reasoningItem);
-            }
-            // ai content
-            let { content } = lcMsg;
-            if (additional_kwargs?.refusal) {
-                if (typeof content === "string") {
-                    content = [{ type: "output_text", text: content, annotations: [] }];
-                }
-                content = [
-                    ...content,
-                    { type: "refusal", refusal: additional_kwargs.refusal },
-                ];
-            }
-            input.push({
-                type: "message",
-                role: "assistant",
-                ...(lcMsg.id && !zdrEnabled ? { id: lcMsg.id } : {}),
-                content: typeof content === "string"
-                    ? content
-                    : content.flatMap((item) => {
-                        if (item.type === "text") {
-                            return {
-                                type: "output_text",
-                                text: item.text,
-                                // @ts-expect-error TODO: add types for `annotations`
-                                annotations: item.annotations ?? [],
-                            };
-                        }
-                        if (item.type === "output_text" || item.type === "refusal") {
-                            return item;
-                        }
-                        return [];
-                    }),
-            });
-            const functionCallIds = additional_kwargs?.[_FUNCTION_CALL_IDS_MAP_KEY];
-            if (isAIMessage(lcMsg) && !!lcMsg.tool_calls?.length) {
-                input.push(...lcMsg.tool_calls.map((toolCall) => ({
-                    type: "function_call",
-                    name: toolCall.name,
-                    arguments: JSON.stringify(toolCall.args),
-                    call_id: toolCall.id,
-                    ...(zdrEnabled ? { id: functionCallIds?.[toolCall.id] } : {}),
-                })));
-            }
-            else if (additional_kwargs?.tool_calls) {
-                input.push(...additional_kwargs.tool_calls.map((toolCall) => ({
-                    type: "function_call",
-                    name: toolCall.function.name,
-                    call_id: toolCall.id,
-                    arguments: toolCall.function.arguments,
-                    ...(zdrEnabled ? { id: functionCallIds?.[toolCall.id] } : {}),
-                })));
-            }
-            const toolOutputs = lcMsg.response_metadata.output?.length
-                ? lcMsg.response_metadata.output
-                : additional_kwargs.tool_outputs;
-            const fallthroughCallTypes = [
-                "computer_call",
-                "mcp_call",
-                "code_interpreter_call",
-                "image_generation_call",
+            return [
+                completionParam,
+                audioMessage,
             ];
-            if (toolOutputs != null) {
-                const castToolOutputs = toolOutputs;
-                const fallthroughCalls = castToolOutputs?.filter((item) => fallthroughCallTypes.includes(item.type));
-                if (fallthroughCalls.length > 0)
-                    input.push(...fallthroughCalls);
-            }
-            return input;
-        }
-        if (role === "user" || role === "system" || role === "developer") {
-            if (typeof lcMsg.content === "string") {
-                return { type: "message", role, content: lcMsg.content };
-            }
-            const messages = [];
-            const content = lcMsg.content.flatMap((item) => {
-                if (item.type === "mcp_approval_response") {
-                    messages.push({
-                        type: "mcp_approval_response",
-                        approval_request_id: item.approval_request_id,
-                        approve: item.approve,
-                    });
-                }
-                if (isDataContentBlock(item)) {
-                    return convertToProviderContentBlock(item, completionsApiContentBlockConverter);
-                }
-                if (item.type === "text") {
-                    return {
-                        type: "input_text",
-                        text: item.text,
-                    };
-                }
-                if (item.type === "image_url") {
-                    return {
-                        type: "input_image",
-                        image_url: typeof item.image_url === "string"
-                            ? item.image_url
-                            : item.image_url.url,
-                        detail: typeof item.image_url === "string"
-                            ? "auto"
-                            : item.image_url.detail,
-                    };
-                }
-                if (item.type === "input_text" ||
-                    item.type === "input_image" ||
-                    item.type === "input_file") {
-                    return item;
-                }
-                return [];
-            });
-            if (content.length > 0) {
-                messages.push({ type: "message", role, content });
-            }
-            return messages;
-        }
-        console.warn(`Unsupported role found when converting to OpenAI Responses API: ${role}`);
-        return [];
-    });
-}
-function _convertOpenAIResponsesMessageToBaseMessage(response) {
-    if (response.error) {
-        // TODO: add support for `addLangChainErrorFields`
-        const error = new Error(response.error.message);
-        error.name = response.error.code;
-        throw error;
-    }
-    let messageId;
-    const content = [];
-    const tool_calls = [];
-    const invalid_tool_calls = [];
-    const response_metadata = {
-        model: response.model,
-        created_at: response.created_at,
-        id: response.id,
-        incomplete_details: response.incomplete_details,
-        metadata: response.metadata,
-        object: response.object,
-        status: response.status,
-        user: response.user,
-        // for compatibility with chat completion calls.
-        model_name: response.model,
-    };
-    const additional_kwargs = {};
-    for (const item of response.output) {
-        if (item.type === "message") {
-            messageId = item.id;
-            content.push(...item.content.flatMap((part) => {
-                if (part.type === "output_text") {
-                    if ("parsed" in part && part.parsed != null) {
-                        additional_kwargs.parsed = part.parsed;
-                    }
-                    return {
-                        type: "text",
-                        text: part.text,
-                        annotations: part.annotations,
-                    };
-                }
-                if (part.type === "refusal") {
-                    additional_kwargs.refusal = part.refusal;
-                    return [];
-                }
-                return part;
-            }));
-        }
-        else if (item.type === "function_call") {
-            const fnAdapter = {
-                function: { name: item.name, arguments: item.arguments },
-                id: item.call_id,
-            };
-            try {
-                tool_calls.push(parseToolCall(fnAdapter, { returnId: true }));
-            }
-            catch (e) {
-                let errMessage;
-                if (typeof e === "object" &&
-                    e != null &&
-                    "message" in e &&
-                    typeof e.message === "string") {
-                    errMessage = e.message;
-                }
-                invalid_tool_calls.push(makeInvalidToolCall(fnAdapter, errMessage));
-            }
-            additional_kwargs[_FUNCTION_CALL_IDS_MAP_KEY] ??= {};
-            if (item.id) {
-                additional_kwargs[_FUNCTION_CALL_IDS_MAP_KEY][item.call_id] = item.id;
-            }
-        }
-        else if (item.type === "reasoning") {
-            additional_kwargs.reasoning = item;
         }
-        else {
-            additional_kwargs.tool_outputs ??= [];
-            additional_kwargs.tool_outputs.push(item);
-        }
-    }
-    return new AIMessage({
-        id: messageId,
-        content,
-        tool_calls,
-        invalid_tool_calls,
-        usage_metadata: response.usage,
-        additional_kwargs,
-        response_metadata,
+        return completionParam;
     });
 }
-function _convertOpenAIResponsesDeltaToBaseMessageChunk(chunk) {
-    const content = [];
-    let generationInfo = {};
-    let usage_metadata;
-    const tool_call_chunks = [];
-    const response_metadata = {};
-    const additional_kwargs = {};
-    let id;
-    if (chunk.type === "response.output_text.delta") {
-        content.push({
-            type: "text",
-            text: chunk.delta,
-            index: chunk.content_index,
-        });
+/** @internal */
+class BaseChatOpenAI extends BaseChatModel {
+    _llmType() {
+        return "openai";
     }
-    else if (chunk.type === "response.output_text_annotation.added") {
-        content.push({
-            type: "text",
-            text: "",
-            annotations: [chunk.annotation],
-            index: chunk.content_index,
-        });
+    static lc_name() {
+        return "ChatOpenAI";
     }
-    else if (chunk.type === "response.output_item.added" &&
-        chunk.item.type === "message") {
-        id = chunk.item.id;
+    get callKeys() {
+        return [
+            ...super.callKeys,
+            "options",
+            "function_call",
+            "functions",
+            "tools",
+            "tool_choice",
+            "promptIndex",
+            "response_format",
+            "seed",
+            "reasoning",
+            "service_tier",
+        ];
     }
-    else if (chunk.type === "response.output_item.added" &&
-        chunk.item.type === "function_call") {
-        tool_call_chunks.push({
-            type: "tool_call_chunk",
-            name: chunk.item.name,
-            args: chunk.item.arguments,
-            id: chunk.item.call_id,
-            index: chunk.output_index,
-        });
-        additional_kwargs[_FUNCTION_CALL_IDS_MAP_KEY] = {
-            [chunk.item.call_id]: chunk.item.id,
+    get lc_secrets() {
+        return {
+            apiKey: "OPENAI_API_KEY",
+            organization: "OPENAI_ORGANIZATION",
         };
     }
-    else if (chunk.type === "response.output_item.done" &&
-        [
-            "web_search_call",
-            "file_search_call",
-            "computer_call",
-            "code_interpreter_call",
-            "mcp_call",
-            "mcp_list_tools",
-            "mcp_approval_request",
-            "image_generation_call",
-        ].includes(chunk.item.type)) {
-        additional_kwargs.tool_outputs = [chunk.item];
-    }
-    else if (chunk.type === "response.created") {
-        response_metadata.id = chunk.response.id;
-        response_metadata.model_name = chunk.response.model;
-        response_metadata.model = chunk.response.model;
-    }
-    else if (chunk.type === "response.completed") {
-        const msg = _convertOpenAIResponsesMessageToBaseMessage(chunk.response);
-        usage_metadata = chunk.response.usage;
-        if (chunk.response.text?.format?.type === "json_schema") {
-            additional_kwargs.parsed ??= JSON.parse(msg.text);
-        }
-        for (const [key, value] of Object.entries(chunk.response)) {
-            if (key !== "id")
-                response_metadata[key] = value;
-        }
-    }
-    else if (chunk.type === "response.function_call_arguments.delta") {
-        tool_call_chunks.push({
-            type: "tool_call_chunk",
-            args: chunk.delta,
-            index: chunk.output_index,
-        });
-    }
-    else if (chunk.type === "response.web_search_call.completed" ||
-        chunk.type === "response.file_search_call.completed") {
-        generationInfo = {
-            tool_outputs: {
-                id: chunk.item_id,
-                type: chunk.type.replace("response.", "").replace(".completed", ""),
-                status: "completed",
-            },
+    get lc_aliases() {
+        return {
+            apiKey: "openai_api_key",
         };
     }
-    else if (chunk.type === "response.refusal.done") {
-        additional_kwargs.refusal = chunk.refusal;
-    }
-    else if (chunk.type === "response.output_item.added" &&
-        "item" in chunk &&
-        chunk.item.type === "reasoning") {
-        const summary = chunk
-            .item.summary
-            ? chunk.item.summary.map((s, index) => ({
-                ...s,
-                index,
-            }))
-            : undefined;
-        additional_kwargs.reasoning = {
-            // We only capture ID in the first chunk or else the concatenated result of all chunks will
-            // have an ID field that is repeated once per chunk. There is special handling for the `type`
-            // field that prevents this, however.
-            id: chunk.item.id,
-            type: chunk.item.type,
-            ...(summary ? { summary } : {}),
-        };
+    get lc_serializable_keys() {
+        return [
+            "configuration",
+            "logprobs",
+            "topLogprobs",
+            "prefixMessages",
+            "supportsStrictToolCalling",
+            "modalities",
+            "audio",
+            "temperature",
+            "maxTokens",
+            "topP",
+            "frequencyPenalty",
+            "presencePenalty",
+            "n",
+            "logitBias",
+            "user",
+            "streaming",
+            "streamUsage",
+            "model",
+            "modelKwargs",
+            "stop",
+            "stopSequences",
+            "timeout",
+            "apiKey",
+            "cache",
+            "maxConcurrency",
+            "maxRetries",
+            "verbose",
+            "callbacks",
+            "tags",
+            "metadata",
+            "disableStreaming",
+            "zdrEnabled",
+            "reasoning",
+        ];
     }
-    else if (chunk.type === "response.reasoning_summary_part.added") {
-        additional_kwargs.reasoning = {
-            type: "reasoning",
-            summary: [{ ...chunk.part, index: chunk.summary_index }],
+    getLsParams(options) {
+        const params = this.invocationParams(options);
+        return {
+            ls_provider: "openai",
+            ls_model_name: this.model,
+            ls_model_type: "chat",
+            ls_temperature: params.temperature ?? undefined,
+            ls_max_tokens: params.max_tokens ?? undefined,
+            ls_stop: options.stop,
         };
     }
-    else if (chunk.type === "response.reasoning_summary_text.delta") {
-        additional_kwargs.reasoning = {
-            type: "reasoning",
-            summary: [
-                { text: chunk.delta, type: "summary_text", index: chunk.summary_index },
-            ],
+    /** @ignore */
+    _identifyingParams() {
+        return {
+            model_name: this.model,
+            ...this.invocationParams(),
+            ...this.clientConfig,
         };
     }
-    else if (chunk.type === "response.image_generation_call.partial_image") {
-        // noop/fixme: retaining partial images in a message chunk means that _all_
-        // partial images get kept in history, so we don't do anything here.
-        return null;
-    }
-    else {
-        return null;
-    }
-    return new ChatGenerationChunk({
-        // Legacy reasons, `onLLMNewToken` should pulls this out
-        text: content.map((part) => part.text).join(""),
-        message: new AIMessageChunk({
-            id,
-            content,
-            tool_call_chunks,
-            usage_metadata,
-            additional_kwargs,
-            response_metadata,
-        }),
-        generationInfo,
-    });
-}
-function isBuiltInTool(tool) {
-    return "type" in tool && tool.type !== "function";
-}
-function isBuiltInToolChoice(tool_choice) {
-    return (tool_choice != null &&
-        typeof tool_choice === "object" &&
-        "type" in tool_choice &&
-        tool_choice.type !== "function");
-}
-function _reduceChatOpenAIToolsForResponses(tools, fields) {
-    const reducedTools = [];
-    for (const tool of tools) {
-        if (isBuiltInTool(tool)) {
-            if (tool.type === "image_generation" && fields?.stream) {
-                // OpenAI sends a 400 error if partial_images is not set and we want to stream.
-                // We also set it to 1 since we don't support partial images yet.
-                tool.partial_images = 1;
-            }
-            reducedTools.push(tool);
-        }
-        else if (isOpenAITool(tool)) {
-            reducedTools.push({
-                type: "function",
-                name: tool.function.name,
-                parameters: tool.function.parameters,
-                description: tool.function.description,
-                strict: fields?.strict ?? null,
-            });
-        }
-    }
-    return reducedTools;
-}
-function _convertChatOpenAIToolTypeToOpenAITool(tool, fields) {
-    if (isOpenAITool(tool)) {
-        if (fields?.strict !== undefined) {
-            return {
-                ...tool,
-                function: {
-                    ...tool.function,
-                    strict: fields.strict,
-                },
-            };
-        }
-        return tool;
-    }
-    return _convertToOpenAITool(tool, fields);
-}
-function isReasoningModel(model) {
-    return model && /^o\d/.test(model);
-}
-/**
- * OpenAI chat model integration.
- *
- * To use with Azure, import the `AzureChatOpenAI` class.
- *
- * Setup:
- * Install `@langchain/openai` and set an environment variable named `OPENAI_API_KEY`.
- *
- * ```bash
- * npm install @langchain/openai
- * export OPENAI_API_KEY="your-api-key"
- * ```
- *
- * ## [Constructor args](https://api.js.langchain.com/classes/langchain_openai.ChatOpenAI.html#constructor)
- *
- * ## [Runtime args](https://api.js.langchain.com/interfaces/langchain_openai.ChatOpenAICallOptions.html)
- *
- * Runtime args can be passed as the second argument to any of the base runnable methods `.invoke`. `.stream`, `.batch`, etc.
- * They can also be passed via `.withConfig`, or the second arg in `.bindTools`, like shown in the examples below:
- *
- * ```typescript
- * // When calling `.withConfig`, call options should be passed via the first argument
- * const llmWithArgsBound = llm.withConfig({
- *   stop: ["\n"],
- *   tools: [...],
- * });
- *
- * // When calling `.bindTools`, call options should be passed via the second argument
- * const llmWithTools = llm.bindTools(
- *   [...],
- *   {
- *     tool_choice: "auto",
- *   }
- * );
- * ```
- *
- * ## Examples
- *
- * <details open>
- * <summary><strong>Instantiate</strong></summary>
- *
- * ```typescript
- * import { ChatOpenAI } from '@langchain/openai';
- *
- * const llm = new ChatOpenAI({
- *   model: "gpt-4o",
- *   temperature: 0,
- *   maxTokens: undefined,
- *   timeout: undefined,
- *   maxRetries: 2,
- *   // apiKey: "...",
- *   // baseUrl: "...",
- *   // organization: "...",
- *   // other params...
- * });
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Invoking</strong></summary>
- *
- * ```typescript
- * const input = `Translate "I love programming" into French.`;
- *
- * // Models also accept a list of chat messages or a formatted prompt
- * const result = await llm.invoke(input);
- * console.log(result);
- * ```
- *
- * ```txt
- * AIMessage {
- *   "id": "chatcmpl-9u4Mpu44CbPjwYFkTbeoZgvzB00Tz",
- *   "content": "J'adore la programmation.",
- *   "response_metadata": {
- *     "tokenUsage": {
- *       "completionTokens": 5,
- *       "promptTokens": 28,
- *       "totalTokens": 33
- *     },
- *     "finish_reason": "stop",
- *     "system_fingerprint": "fp_3aa7262c27"
- *   },
- *   "usage_metadata": {
- *     "input_tokens": 28,
- *     "output_tokens": 5,
- *     "total_tokens": 33
- *   }
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Streaming Chunks</strong></summary>
- *
- * ```typescript
- * for await (const chunk of await llm.stream(input)) {
- *   console.log(chunk);
- * }
- * ```
- *
- * ```txt
- * AIMessageChunk {
- *   "id": "chatcmpl-9u4NWB7yUeHCKdLr6jP3HpaOYHTqs",
- *   "content": ""
- * }
- * AIMessageChunk {
- *   "content": "J"
- * }
- * AIMessageChunk {
- *   "content": "'adore"
- * }
- * AIMessageChunk {
- *   "content": " la"
- * }
- * AIMessageChunk {
- *   "content": " programmation",,
- * }
- * AIMessageChunk {
- *   "content": ".",,
- * }
- * AIMessageChunk {
- *   "content": "",
- *   "response_metadata": {
- *     "finish_reason": "stop",
- *     "system_fingerprint": "fp_c9aa9c0491"
- *   },
- * }
- * AIMessageChunk {
- *   "content": "",
- *   "usage_metadata": {
- *     "input_tokens": 28,
- *     "output_tokens": 5,
- *     "total_tokens": 33
- *   }
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Aggregate Streamed Chunks</strong></summary>
- *
- * ```typescript
- * import { AIMessageChunk } from '@langchain/core/messages';
- * import { concat } from '@langchain/core/utils/stream';
- *
- * const stream = await llm.stream(input);
- * let full: AIMessageChunk | undefined;
- * for await (const chunk of stream) {
- *   full = !full ? chunk : concat(full, chunk);
- * }
- * console.log(full);
- * ```
- *
- * ```txt
- * AIMessageChunk {
- *   "id": "chatcmpl-9u4PnX6Fy7OmK46DASy0bH6cxn5Xu",
- *   "content": "J'adore la programmation.",
- *   "response_metadata": {
- *     "prompt": 0,
- *     "completion": 0,
- *     "finish_reason": "stop",
- *   },
- *   "usage_metadata": {
- *     "input_tokens": 28,
- *     "output_tokens": 5,
- *     "total_tokens": 33
- *   }
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Bind tools</strong></summary>
- *
- * ```typescript
- * import { z } from 'zod';
- *
- * const GetWeather = {
- *   name: "GetWeather",
- *   description: "Get the current weather in a given location",
- *   schema: z.object({
- *     location: z.string().describe("The city and state, e.g. San Francisco, CA")
- *   }),
- * }
- *
- * const GetPopulation = {
- *   name: "GetPopulation",
- *   description: "Get the current population in a given location",
- *   schema: z.object({
- *     location: z.string().describe("The city and state, e.g. San Francisco, CA")
- *   }),
- * }
- *
- * const llmWithTools = llm.bindTools(
- *   [GetWeather, GetPopulation],
- *   {
- *     // strict: true  // enforce tool args schema is respected
- *   }
- * );
- * const aiMsg = await llmWithTools.invoke(
- *   "Which city is hotter today and which is bigger: LA or NY?"
- * );
- * console.log(aiMsg.tool_calls);
- * ```
- *
- * ```txt
- * [
- *   {
- *     name: 'GetWeather',
- *     args: { location: 'Los Angeles, CA' },
- *     type: 'tool_call',
- *     id: 'call_uPU4FiFzoKAtMxfmPnfQL6UK'
- *   },
- *   {
- *     name: 'GetWeather',
- *     args: { location: 'New York, NY' },
- *     type: 'tool_call',
- *     id: 'call_UNkEwuQsHrGYqgDQuH9nPAtX'
- *   },
- *   {
- *     name: 'GetPopulation',
- *     args: { location: 'Los Angeles, CA' },
- *     type: 'tool_call',
- *     id: 'call_kL3OXxaq9OjIKqRTpvjaCH14'
- *   },
- *   {
- *     name: 'GetPopulation',
- *     args: { location: 'New York, NY' },
- *     type: 'tool_call',
- *     id: 'call_s9KQB1UWj45LLGaEnjz0179q'
- *   }
- * ]
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Structured Output</strong></summary>
- *
- * ```typescript
- * import { z } from 'zod';
- *
- * const Joke = z.object({
- *   setup: z.string().describe("The setup of the joke"),
- *   punchline: z.string().describe("The punchline to the joke"),
- *   rating: z.number().nullable().describe("How funny the joke is, from 1 to 10")
- * }).describe('Joke to tell user.');
- *
- * const structuredLlm = llm.withStructuredOutput(Joke, {
- *   name: "Joke",
- *   strict: true, // Optionally enable OpenAI structured outputs
- * });
- * const jokeResult = await structuredLlm.invoke("Tell me a joke about cats");
- * console.log(jokeResult);
- * ```
- *
- * ```txt
- * {
- *   setup: 'Why was the cat sitting on the computer?',
- *   punchline: 'Because it wanted to keep an eye on the mouse!',
- *   rating: 7
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>JSON Object Response Format</strong></summary>
- *
- * ```typescript
- * const jsonLlm = llm.withConfig({ response_format: { type: "json_object" } });
- * const jsonLlmAiMsg = await jsonLlm.invoke(
- *   "Return a JSON object with key 'randomInts' and a value of 10 random ints in [0-99]"
- * );
- * console.log(jsonLlmAiMsg.content);
- * ```
- *
- * ```txt
- * {
- *   "randomInts": [23, 87, 45, 12, 78, 34, 56, 90, 11, 67]
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Multimodal</strong></summary>
- *
- * ```typescript
- * import { HumanMessage } from '@langchain/core/messages';
- *
- * const imageUrl = "https://example.com/image.jpg";
- * const imageData = await fetch(imageUrl).then(res => res.arrayBuffer());
- * const base64Image = Buffer.from(imageData).toString('base64');
- *
- * const message = new HumanMessage({
- *   content: [
- *     { type: "text", text: "describe the weather in this image" },
- *     {
- *       type: "image_url",
- *       image_url: { url: `data:image/jpeg;base64,${base64Image}` },
- *     },
- *   ]
- * });
- *
- * const imageDescriptionAiMsg = await llm.invoke([message]);
- * console.log(imageDescriptionAiMsg.content);
- * ```
- *
- * ```txt
- * The weather in the image appears to be clear and sunny. The sky is mostly blue with a few scattered white clouds, indicating fair weather. The bright sunlight is casting shadows on the green, grassy hill, suggesting it is a pleasant day with good visibility. There are no signs of rain or stormy conditions.
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Usage Metadata</strong></summary>
- *
- * ```typescript
- * const aiMsgForMetadata = await llm.invoke(input);
- * console.log(aiMsgForMetadata.usage_metadata);
- * ```
- *
- * ```txt
- * { input_tokens: 28, output_tokens: 5, total_tokens: 33 }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Logprobs</strong></summary>
- *
- * ```typescript
- * const logprobsLlm = new ChatOpenAI({ logprobs: true });
- * const aiMsgForLogprobs = await logprobsLlm.invoke(input);
- * console.log(aiMsgForLogprobs.response_metadata.logprobs);
- * ```
- *
- * ```txt
- * {
- *   content: [
- *     {
- *       token: 'J',
- *       logprob: -0.000050616763,
- *       bytes: [Array],
- *       top_logprobs: []
- *     },
- *     {
- *       token: "'",
- *       logprob: -0.01868736,
- *       bytes: [Array],
- *       top_logprobs: []
- *     },
- *     {
- *       token: 'ad',
- *       logprob: -0.0000030545007,
- *       bytes: [Array],
- *       top_logprobs: []
- *     },
- *     { token: 'ore', logprob: 0, bytes: [Array], top_logprobs: [] },
- *     {
- *       token: ' la',
- *       logprob: -0.515404,
- *       bytes: [Array],
- *       top_logprobs: []
- *     },
- *     {
- *       token: ' programm',
- *       logprob: -0.0000118755715,
- *       bytes: [Array],
- *       top_logprobs: []
- *     },
- *     { token: 'ation', logprob: 0, bytes: [Array], top_logprobs: [] },
- *     {
- *       token: '.',
- *       logprob: -0.0000037697225,
- *       bytes: [Array],
- *       top_logprobs: []
- *     }
- *   ],
- *   refusal: null
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Response Metadata</strong></summary>
- *
- * ```typescript
- * const aiMsgForResponseMetadata = await llm.invoke(input);
- * console.log(aiMsgForResponseMetadata.response_metadata);
- * ```
- *
- * ```txt
- * {
- *   tokenUsage: { completionTokens: 5, promptTokens: 28, totalTokens: 33 },
- *   finish_reason: 'stop',
- *   system_fingerprint: 'fp_3aa7262c27'
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>JSON Schema Structured Output</strong></summary>
- *
- * ```typescript
- * const llmForJsonSchema = new ChatOpenAI({
- *   model: "gpt-4o-2024-08-06",
- * }).withStructuredOutput(
- *   z.object({
- *     command: z.string().describe("The command to execute"),
- *     expectedOutput: z.string().describe("The expected output of the command"),
- *     options: z
- *       .array(z.string())
- *       .describe("The options you can pass to the command"),
- *   }),
- *   {
- *     method: "jsonSchema",
- *     strict: true, // Optional when using the `jsonSchema` method
- *   }
- * );
- *
- * const jsonSchemaRes = await llmForJsonSchema.invoke(
- *   "What is the command to list files in a directory?"
- * );
- * console.log(jsonSchemaRes);
- * ```
- *
- * ```txt
- * {
- *   command: 'ls',
- *   expectedOutput: 'A list of files and subdirectories within the specified directory.',
- *   options: [
- *     '-a: include directory entries whose names begin with a dot (.).',
- *     '-l: use a long listing format.',
- *     '-h: with -l, print sizes in human readable format (e.g., 1K, 234M, 2G).',
- *     '-t: sort by time, newest first.',
- *     '-r: reverse order while sorting.',
- *     '-S: sort by file size, largest first.',
- *     '-R: list subdirectories recursively.'
- *   ]
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Audio Outputs</strong></summary>
- *
- * ```typescript
- * import { ChatOpenAI } from "@langchain/openai";
- *
- * const modelWithAudioOutput = new ChatOpenAI({
- *   model: "gpt-4o-audio-preview",
- *   // You may also pass these fields to `.withConfig` as a call argument.
- *   modalities: ["text", "audio"], // Specifies that the model should output audio.
- *   audio: {
- *     voice: "alloy",
- *     format: "wav",
- *   },
- * });
- *
- * const audioOutputResult = await modelWithAudioOutput.invoke("Tell me a joke about cats.");
- * const castMessageContent = audioOutputResult.content[0] as Record<string, any>;
- *
- * console.log({
- *   ...castMessageContent,
- *   data: castMessageContent.data.slice(0, 100) // Sliced for brevity
- * })
- * ```
- *
- * ```txt
- * {
- *   id: 'audio_67117718c6008190a3afad3e3054b9b6',
- *   data: 'UklGRqYwBgBXQVZFZm10IBAAAAABAAEAwF0AAIC7AAACABAATElTVBoAAABJTkZPSVNGVA4AAABMYXZmNTguMjkuMTAwAGRhdGFg',
- *   expires_at: 1729201448,
- *   transcript: 'Sure! Why did the cat sit on the computer? Because it wanted to keep an eye on the mouse!'
- * }
- * ```
- * </details>
- *
- * <br />
- *
- * <details>
- * <summary><strong>Audio Outputs</strong></summary>
- *
- * ```typescript
- * import { ChatOpenAI } from "@langchain/openai";
- *
- * const modelWithAudioOutput = new ChatOpenAI({
- *   model: "gpt-4o-audio-preview",
- *   // You may also pass these fields to `.withConfig` as a call argument.
- *   modalities: ["text", "audio"], // Specifies that the model should output audio.
- *   audio: {
- *     voice: "alloy",
- *     format: "wav",
- *   },
- * });
- *
- * const audioOutputResult = await modelWithAudioOutput.invoke("Tell me a joke about cats.");
- * const castAudioContent = audioOutputResult.additional_kwargs.audio as Record<string, any>;
- *
- * console.log({
- *   ...castAudioContent,
- *   data: castAudioContent.data.slice(0, 100) // Sliced for brevity
- * })
- * ```
- *
- * ```txt
- * {
- *   id: 'audio_67117718c6008190a3afad3e3054b9b6',
- *   data: 'UklGRqYwBgBXQVZFZm10IBAAAAABAAEAwF0AAIC7AAACABAATElTVBoAAABJTkZPSVNGVA4AAABMYXZmNTguMjkuMTAwAGRhdGFg',
- *   expires_at: 1729201448,
- *   transcript: 'Sure! Why did the cat sit on the computer? Because it wanted to keep an eye on the mouse!'
- * }
- * ```
- * </details>
- *
- * <br />
- */
-export class ChatOpenAI extends BaseChatModel {
-    static lc_name() {
-        return "ChatOpenAI";
-    }
-    get callKeys() {
-        return [
-            ...super.callKeys,
-            "options",
-            "function_call",
-            "functions",
-            "tools",
-            "tool_choice",
-            "promptIndex",
-            "response_format",
-            "seed",
-            "reasoning_effort",
-        ];
-    }
-    get lc_secrets() {
-        return {
-            openAIApiKey: "OPENAI_API_KEY",
-            apiKey: "OPENAI_API_KEY",
-            organization: "OPENAI_ORGANIZATION",
-        };
-    }
-    get lc_aliases() {
-        return {
-            modelName: "model",
-            openAIApiKey: "openai_api_key",
-            apiKey: "openai_api_key",
-        };
-    }
-    get lc_serializable_keys() {
-        return [
-            "configuration",
-            "logprobs",
-            "topLogprobs",
-            "prefixMessages",
-            "supportsStrictToolCalling",
-            "modalities",
-            "audio",
-            "reasoningEffort",
-            "temperature",
-            "maxTokens",
-            "topP",
-            "frequencyPenalty",
-            "presencePenalty",
-            "n",
-            "logitBias",
-            "user",
-            "streaming",
-            "streamUsage",
-            "modelName",
-            "model",
-            "modelKwargs",
-            "stop",
-            "stopSequences",
-            "timeout",
-            "openAIApiKey",
-            "apiKey",
-            "cache",
-            "maxConcurrency",
-            "maxRetries",
-            "verbose",
-            "callbacks",
-            "tags",
-            "metadata",
-            "disableStreaming",
-            "useResponsesApi",
-            "zdrEnabled",
-            "reasoning",
-        ];
+    /**
+     * Get the identifying parameters for the model
+     */
+    identifyingParams() {
+        return this._identifyingParams();
     }
     constructor(fields) {
         super(fields ?? {});
-        Object.defineProperty(this, "lc_serializable", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: true
-        });
         Object.defineProperty(this, "temperature", {
             enumerable: true,
             configurable: true,
@@ -1367,13 +385,6 @@ export class ChatOpenAI extends BaseChatModel {
             writable: true,
             value: void 0
         });
-        /** @deprecated Use "model" instead */
-        Object.defineProperty(this, "modelName", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
         Object.defineProperty(this, "model", {
             enumerable: true,
             configurable: true,
@@ -1440,12 +451,6 @@ export class ChatOpenAI extends BaseChatModel {
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "openAIApiKey", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
         Object.defineProperty(this, "apiKey", {
             enumerable: true,
             configurable: true,
@@ -1498,175 +503,1104 @@ export class ChatOpenAI extends BaseChatModel {
             writable: true,
             value: void 0
         });
-        /**
-         * @deprecated Use {@link reasoning} object instead.
-         */
-        Object.defineProperty(this, "reasoningEffort", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
+        Object.defineProperty(this, "reasoning", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        /**
+         * Must be set to `true` in tenancies with Zero Data Retention. Setting to `true` will disable
+         * output storage in the Responses API, but this DOES NOT enable Zero Data Retention in your
+         * OpenAI organization or project. This must be configured directly with OpenAI.
+         *
+         * See:
+         * https://help.openai.com/en/articles/10503543-data-residency-for-the-openai-api
+         * https://platform.openai.com/docs/api-reference/responses/create#responses-create-store
+         *
+         * @default false
+         */
+        Object.defineProperty(this, "zdrEnabled", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        /**
+         * Service tier to use for this request. Can be "auto", "default", or "flex" or "priority".
+         * Specifies the service tier for prioritization and latency optimization.
+         */
+        Object.defineProperty(this, "service_tier", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "lc_serializable", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: true
+        });
+        this.apiKey =
+            fields?.apiKey ??
+                fields?.configuration?.apiKey ??
+                getEnvironmentVariable("OPENAI_API_KEY");
+        this.organization =
+            fields?.configuration?.organization ??
+                getEnvironmentVariable("OPENAI_ORGANIZATION");
+        this.model = fields?.model ?? this.model;
+        this.modelKwargs = fields?.modelKwargs ?? {};
+        this.timeout = fields?.timeout;
+        this.temperature = fields?.temperature ?? this.temperature;
+        this.topP = fields?.topP ?? this.topP;
+        this.frequencyPenalty = fields?.frequencyPenalty ?? this.frequencyPenalty;
+        this.presencePenalty = fields?.presencePenalty ?? this.presencePenalty;
+        this.logprobs = fields?.logprobs;
+        this.topLogprobs = fields?.topLogprobs;
+        this.n = fields?.n ?? this.n;
+        this.logitBias = fields?.logitBias;
+        this.stop = fields?.stopSequences ?? fields?.stop;
+        this.stopSequences = this.stop;
+        this.user = fields?.user;
+        this.__includeRawResponse = fields?.__includeRawResponse;
+        this.audio = fields?.audio;
+        this.modalities = fields?.modalities;
+        this.reasoning = fields?.reasoning;
+        this.maxTokens = fields?.maxCompletionTokens ?? fields?.maxTokens;
+        this.disableStreaming = fields?.disableStreaming ?? this.disableStreaming;
+        this.streaming = fields?.streaming ?? false;
+        if (this.disableStreaming)
+            this.streaming = false;
+        this.streamUsage = fields?.streamUsage ?? this.streamUsage;
+        if (this.disableStreaming)
+            this.streamUsage = false;
+        this.clientConfig = {
+            apiKey: this.apiKey,
+            organization: this.organization,
+            dangerouslyAllowBrowser: true,
+            ...fields?.configuration,
+        };
+        // If `supportsStrictToolCalling` is explicitly set, use that value.
+        // Else leave undefined so it's not passed to OpenAI.
+        if (fields?.supportsStrictToolCalling !== undefined) {
+            this.supportsStrictToolCalling = fields.supportsStrictToolCalling;
+        }
+        if (fields?.service_tier !== undefined) {
+            this.service_tier = fields.service_tier;
+        }
+        this.zdrEnabled = fields?.zdrEnabled ?? false;
+    }
+    /**
+     * Returns backwards compatible reasoning parameters from constructor params and call options
+     * @internal
+     */
+    _getReasoningParams(options) {
+        if (!isReasoningModel(this.model)) {
+            return;
+        }
+        // apply options in reverse order of importance -- newer options supersede older options
+        let reasoning;
+        if (this.reasoning !== undefined) {
+            reasoning = {
+                ...reasoning,
+                ...this.reasoning,
+            };
+        }
+        if (options?.reasoning !== undefined) {
+            reasoning = {
+                ...reasoning,
+                ...options.reasoning,
+            };
+        }
+        return reasoning;
+    }
+    /**
+     * Returns an openai compatible response format from a set of options
+     * @internal
+     */
+    _getResponseFormat(resFormat) {
+        if (resFormat &&
+            resFormat.type === "json_schema" &&
+            resFormat.json_schema.schema &&
+            isInteropZodSchema(resFormat.json_schema.schema)) {
+            return interopZodResponseFormat(resFormat.json_schema.schema, resFormat.json_schema.name, {
+                description: resFormat.json_schema.description,
+            });
+        }
+        return resFormat;
+    }
+    _getClientOptions(options) {
+        if (!this.client) {
+            const openAIEndpointConfig = {
+                baseURL: this.clientConfig.baseURL,
+            };
+            const endpoint = getEndpoint(openAIEndpointConfig);
+            const params = {
+                ...this.clientConfig,
+                baseURL: endpoint,
+                timeout: this.timeout,
+                maxRetries: 0,
+            };
+            if (!params.baseURL) {
+                delete params.baseURL;
+            }
+            this.client = new OpenAIClient(params);
+        }
+        const requestOptions = {
+            ...this.clientConfig,
+            ...options,
+        };
+        return requestOptions;
+    }
+    // TODO: move to completions class
+    _convertChatOpenAIToolToCompletionsTool(tool, fields) {
+        if (isOpenAITool(tool)) {
+            if (fields?.strict !== undefined) {
+                return {
+                    ...tool,
+                    function: {
+                        ...tool.function,
+                        strict: fields.strict,
+                    },
+                };
+            }
+            return tool;
+        }
+        return _convertToOpenAITool(tool, fields);
+    }
+    bindTools(tools, kwargs) {
+        let strict;
+        if (kwargs?.strict !== undefined) {
+            strict = kwargs.strict;
+        }
+        else if (this.supportsStrictToolCalling !== undefined) {
+            strict = this.supportsStrictToolCalling;
+        }
+        return this.withConfig({
+            tools: tools.map((tool) => isBuiltInTool(tool)
+                ? tool
+                : this._convertChatOpenAIToolToCompletionsTool(tool, { strict })),
+            ...kwargs,
+        });
+    }
+    /** @ignore */
+    _combineLLMOutput(...llmOutputs) {
+        return llmOutputs.reduce((acc, llmOutput) => {
+            if (llmOutput && llmOutput.tokenUsage) {
+                acc.tokenUsage.completionTokens +=
+                    llmOutput.tokenUsage.completionTokens ?? 0;
+                acc.tokenUsage.promptTokens += llmOutput.tokenUsage.promptTokens ?? 0;
+                acc.tokenUsage.totalTokens += llmOutput.tokenUsage.totalTokens ?? 0;
+            }
+            return acc;
+        }, {
+            tokenUsage: {
+                completionTokens: 0,
+                promptTokens: 0,
+                totalTokens: 0,
+            },
         });
-        /**
-         * Options for reasoning models.
-         */
-        Object.defineProperty(this, "reasoning", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
+    }
+    async getNumTokensFromMessages(messages) {
+        let totalCount = 0;
+        let tokensPerMessage = 0;
+        let tokensPerName = 0;
+        // From: https://github.com/openai/openai-cookbook/blob/main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb
+        if (this.model === "gpt-3.5-turbo-0301") {
+            tokensPerMessage = 4;
+            tokensPerName = -1;
+        }
+        else {
+            tokensPerMessage = 3;
+            tokensPerName = 1;
+        }
+        const countPerMessage = await Promise.all(messages.map(async (message) => {
+            const textCount = await this.getNumTokens(message.content);
+            const roleCount = await this.getNumTokens(messageToOpenAIRole(message));
+            const nameCount = message.name !== undefined
+                ? tokensPerName + (await this.getNumTokens(message.name))
+                : 0;
+            let count = textCount + tokensPerMessage + roleCount + nameCount;
+            // From: https://github.com/hmarr/openai-chat-tokens/blob/main/src/index.ts messageTokenEstimate
+            const openAIMessage = message;
+            if (openAIMessage._getType() === "function") {
+                count -= 2;
+            }
+            if (openAIMessage.additional_kwargs?.function_call) {
+                count += 3;
+            }
+            if (openAIMessage?.additional_kwargs.function_call?.name) {
+                count += await this.getNumTokens(openAIMessage.additional_kwargs.function_call?.name);
+            }
+            if (openAIMessage.additional_kwargs.function_call?.arguments) {
+                try {
+                    count += await this.getNumTokens(
+                    // Remove newlines and spaces
+                    JSON.stringify(JSON.parse(openAIMessage.additional_kwargs.function_call?.arguments)));
+                }
+                catch (error) {
+                    console.error("Error parsing function arguments", error, JSON.stringify(openAIMessage.additional_kwargs.function_call));
+                    count += await this.getNumTokens(openAIMessage.additional_kwargs.function_call?.arguments);
+                }
+            }
+            totalCount += count;
+            return count;
+        }));
+        totalCount += 3; // every reply is primed with <|start|>assistant<|message|>
+        return { totalCount, countPerMessage };
+    }
+    /** @internal */
+    async _getNumTokensFromGenerations(generations) {
+        const generationUsages = await Promise.all(generations.map(async (generation) => {
+            if (generation.message.additional_kwargs?.function_call) {
+                return (await this.getNumTokensFromMessages([generation.message]))
+                    .countPerMessage[0];
+            }
+            else {
+                return await this.getNumTokens(generation.message.content);
+            }
+        }));
+        return generationUsages.reduce((a, b) => a + b, 0);
+    }
+    /** @internal */
+    async _getEstimatedTokenCountFromPrompt(messages, functions, function_call) {
+        // It appears that if functions are present, the first system message is padded with a trailing newline. This
+        // was inferred by trying lots of combinations of messages and functions and seeing what the token counts were.
+        let tokens = (await this.getNumTokensFromMessages(messages)).totalCount;
+        // If there are functions, add the function definitions as they count towards token usage
+        if (functions && function_call !== "auto") {
+            const promptDefinitions = formatFunctionDefinitions(functions);
+            tokens += await this.getNumTokens(promptDefinitions);
+            tokens += 9; // Add nine per completion
+        }
+        // If there's a system message _and_ functions are present, subtract four tokens. I assume this is because
+        // functions typically add a system message, but reuse the first one if it's already there. This offsets
+        // the extra 9 tokens added by the function definitions.
+        if (functions && messages.find((m) => m._getType() === "system")) {
+            tokens -= 4;
+        }
+        // If function_call is 'none', add one token.
+        // If it's a FunctionCall object, add 4 + the number of tokens in the function name.
+        // If it's undefined or 'auto', don't add anything.
+        if (function_call === "none") {
+            tokens += 1;
+        }
+        else if (typeof function_call === "object") {
+            tokens += (await this.getNumTokens(function_call.name)) + 4;
+        }
+        return tokens;
+    }
+    withStructuredOutput(outputSchema, config) {
+        // ):
+        // | Runnable<BaseLanguageModelInput, RunOutput>
+        // | Runnable<
+        //     BaseLanguageModelInput,
+        //     { raw: BaseMessage; parsed: RunOutput }
+        //   > {
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        let schema;
+        let name;
+        let method;
+        let includeRaw;
+        if (isStructuredOutputMethodParams(outputSchema)) {
+            schema = outputSchema.schema;
+            name = outputSchema.name;
+            method = outputSchema.method;
+            includeRaw = outputSchema.includeRaw;
+        }
+        else {
+            schema = outputSchema;
+            name = config?.name;
+            method = config?.method;
+            includeRaw = config?.includeRaw;
+        }
+        let llm;
+        let outputParser;
+        if (config?.strict !== undefined && method === "jsonMode") {
+            throw new Error("Argument `strict` is only supported for `method` = 'function_calling'");
+        }
+        if (!this.model.startsWith("gpt-3") &&
+            !this.model.startsWith("gpt-4-") &&
+            this.model !== "gpt-4") {
+            if (method === undefined) {
+                method = "jsonSchema";
+            }
+        }
+        else if (method === "jsonSchema") {
+            console.warn(`[WARNING]: JSON Schema is not supported for model "${this.model}". Falling back to tool calling.`);
+        }
+        if (method === "jsonMode") {
+            let outputFormatSchema;
+            if (isInteropZodSchema(schema)) {
+                outputParser = StructuredOutputParser.fromZodSchema(schema);
+                outputFormatSchema = toJsonSchema(schema);
+            }
+            else {
+                outputParser = new JsonOutputParser();
+            }
+            llm = this.withConfig({
+                response_format: { type: "json_object" },
+                ls_structured_output_format: {
+                    kwargs: { method: "jsonMode" },
+                    schema: outputFormatSchema,
+                },
+            });
+        }
+        else if (method === "jsonSchema") {
+            llm = this.withConfig({
+                response_format: {
+                    type: "json_schema",
+                    json_schema: {
+                        name: name ?? "extract",
+                        description: getSchemaDescription(schema),
+                        schema,
+                        strict: config?.strict,
+                    },
+                },
+                ls_structured_output_format: {
+                    kwargs: { method: "jsonSchema" },
+                    schema: toJsonSchema(schema),
+                },
+            });
+            if (isInteropZodSchema(schema)) {
+                const altParser = StructuredOutputParser.fromZodSchema(schema);
+                outputParser = RunnableLambda.from((aiMessage) => {
+                    if ("parsed" in aiMessage.additional_kwargs) {
+                        return aiMessage.additional_kwargs.parsed;
+                    }
+                    return altParser;
+                });
+            }
+            else {
+                outputParser = new JsonOutputParser();
+            }
+        }
+        else {
+            let functionName = name ?? "extract";
+            // Is function calling
+            if (isInteropZodSchema(schema)) {
+                const asJsonSchema = toJsonSchema(schema);
+                llm = this.withConfig({
+                    tools: [
+                        {
+                            type: "function",
+                            function: {
+                                name: functionName,
+                                description: asJsonSchema.description,
+                                parameters: asJsonSchema,
+                            },
+                        },
+                    ],
+                    tool_choice: {
+                        type: "function",
+                        function: {
+                            name: functionName,
+                        },
+                    },
+                    ls_structured_output_format: {
+                        kwargs: { method: "functionCalling" },
+                        schema: asJsonSchema,
+                    },
+                    // Do not pass `strict` argument to OpenAI if `config.strict` is undefined
+                    ...(config?.strict !== undefined ? { strict: config.strict } : {}),
+                });
+                outputParser = new JsonOutputKeyToolsParser({
+                    returnSingle: true,
+                    keyName: functionName,
+                    zodSchema: schema,
+                });
+            }
+            else {
+                let openAIFunctionDefinition;
+                if (typeof schema.name === "string" &&
+                    typeof schema.parameters === "object" &&
+                    schema.parameters != null) {
+                    openAIFunctionDefinition = schema;
+                    functionName = schema.name;
+                }
+                else {
+                    functionName = schema.title ?? functionName;
+                    openAIFunctionDefinition = {
+                        name: functionName,
+                        description: schema.description ?? "",
+                        parameters: schema,
+                    };
+                }
+                llm = this.withConfig({
+                    tools: [
+                        {
+                            type: "function",
+                            function: openAIFunctionDefinition,
+                        },
+                    ],
+                    tool_choice: {
+                        type: "function",
+                        function: {
+                            name: functionName,
+                        },
+                    },
+                    ls_structured_output_format: {
+                        kwargs: { method: "functionCalling" },
+                        schema: toJsonSchema(schema),
+                    },
+                    // Do not pass `strict` argument to OpenAI if `config.strict` is undefined
+                    ...(config?.strict !== undefined ? { strict: config.strict } : {}),
+                });
+                outputParser = new JsonOutputKeyToolsParser({
+                    returnSingle: true,
+                    keyName: functionName,
+                });
+            }
+        }
+        if (!includeRaw) {
+            return llm.pipe(outputParser);
+        }
+        const parserAssign = RunnablePassthrough.assign({
+            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+            parsed: (input, config) => outputParser.invoke(input.raw, config),
         });
-        /**
-         * Whether to use the responses API for all requests. If `false` the responses API will be used
-         * only when required in order to fulfill the request.
-         */
-        Object.defineProperty(this, "useResponsesApi", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: false
+        const parserNone = RunnablePassthrough.assign({
+            parsed: () => null,
         });
-        /**
-         * Must be set to `true` in tenancies with Zero Data Retention. Setting to `true` will disable
-         * output storage in the Responses API, but this DOES NOT enable Zero Data Retention in your
-         * OpenAI organization or project. This must be configured directly with OpenAI.
-         *
-         * See:
-         * https://help.openai.com/en/articles/10503543-data-residency-for-the-openai-api
-         * https://platform.openai.com/docs/api-reference/responses/create#responses-create-store
-         *
-         * @default false
-         */
-        Object.defineProperty(this, "zdrEnabled", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
+        const parsedWithFallback = parserAssign.withFallbacks({
+            fallbacks: [parserNone],
         });
-        this.openAIApiKey =
-            fields?.apiKey ??
-                fields?.openAIApiKey ??
-                fields?.configuration?.apiKey ??
-                getEnvironmentVariable("OPENAI_API_KEY");
-        this.apiKey = this.openAIApiKey;
-        this.organization =
-            fields?.configuration?.organization ??
-                getEnvironmentVariable("OPENAI_ORGANIZATION");
-        this.model = fields?.model ?? fields?.modelName ?? this.model;
-        this.modelName = this.model;
-        this.modelKwargs = fields?.modelKwargs ?? {};
-        this.timeout = fields?.timeout;
-        this.temperature = fields?.temperature ?? this.temperature;
-        this.topP = fields?.topP ?? this.topP;
-        this.frequencyPenalty = fields?.frequencyPenalty ?? this.frequencyPenalty;
-        this.presencePenalty = fields?.presencePenalty ?? this.presencePenalty;
-        this.logprobs = fields?.logprobs;
-        this.topLogprobs = fields?.topLogprobs;
-        this.n = fields?.n ?? this.n;
-        this.logitBias = fields?.logitBias;
-        this.stop = fields?.stopSequences ?? fields?.stop;
-        this.stopSequences = this.stop;
-        this.user = fields?.user;
-        this.__includeRawResponse = fields?.__includeRawResponse;
-        this.audio = fields?.audio;
-        this.modalities = fields?.modalities;
-        this.reasoningEffort = fields?.reasoningEffort ?? fields?.reasoning?.effort;
-        this.reasoning =
-            fields?.reasoning ??
-                (fields?.reasoningEffort
-                    ? { effort: fields.reasoningEffort }
-                    : undefined);
-        this.maxTokens = fields?.maxCompletionTokens ?? fields?.maxTokens;
-        this.useResponsesApi = fields?.useResponsesApi ?? this.useResponsesApi;
-        this.disableStreaming = fields?.disableStreaming ?? this.disableStreaming;
-        this.streaming = fields?.streaming ?? false;
-        if (this.disableStreaming)
-            this.streaming = false;
-        this.streamUsage = fields?.streamUsage ?? this.streamUsage;
-        if (this.disableStreaming)
-            this.streamUsage = false;
-        this.clientConfig = {
-            apiKey: this.apiKey,
-            organization: this.organization,
-            dangerouslyAllowBrowser: true,
-            ...fields?.configuration,
-        };
-        // If `supportsStrictToolCalling` is explicitly set, use that value.
-        // Else leave undefined so it's not passed to OpenAI.
-        if (fields?.supportsStrictToolCalling !== undefined) {
-            this.supportsStrictToolCalling = fields.supportsStrictToolCalling;
-        }
-        this.zdrEnabled = fields?.zdrEnabled ?? false;
-    }
-    getLsParams(options) {
-        const params = this.invocationParams(options);
-        return {
-            ls_provider: "openai",
-            ls_model_name: this.model,
-            ls_model_type: "chat",
-            ls_temperature: params.temperature ?? undefined,
-            ls_max_tokens: params.max_tokens ?? undefined,
-            ls_stop: options.stop,
-        };
+        return RunnableSequence.from([{ raw: llm }, parsedWithFallback]);
     }
-    bindTools(tools, kwargs) {
+}
+/**
+ * OpenAI Responses API implementation.
+ *
+ * Will be exported in a later version of @langchain/openai.
+ *
+ * @internal
+ */
+export class ChatOpenAIResponses extends BaseChatOpenAI {
+    invocationParams(options) {
         let strict;
-        if (kwargs?.strict !== undefined) {
-            strict = kwargs.strict;
+        if (options?.strict !== undefined) {
+            strict = options.strict;
         }
         else if (this.supportsStrictToolCalling !== undefined) {
             strict = this.supportsStrictToolCalling;
         }
-        return this.withConfig({
-            tools: tools.map((tool) => isBuiltInTool(tool)
-                ? tool
-                : _convertChatOpenAIToolTypeToOpenAITool(tool, { strict })),
-            ...kwargs,
+        const params = {
+            model: this.model,
+            temperature: this.temperature,
+            top_p: this.topP,
+            user: this.user,
+            // if include_usage is set or streamUsage then stream must be set to true.
+            stream: this.streaming,
+            previous_response_id: options?.previous_response_id,
+            truncation: options?.truncation,
+            include: options?.include,
+            tools: options?.tools?.length
+                ? this._reduceChatOpenAITools(options.tools, {
+                    stream: this.streaming,
+                    strict,
+                })
+                : undefined,
+            tool_choice: isBuiltInToolChoice(options?.tool_choice)
+                ? options?.tool_choice
+                : (() => {
+                    const formatted = formatToOpenAIToolChoice(options?.tool_choice);
+                    if (typeof formatted === "object" && "type" in formatted) {
+                        return { type: "function", name: formatted.function.name };
+                    }
+                    else {
+                        return undefined;
+                    }
+                })(),
+            text: (() => {
+                if (options?.text)
+                    return options.text;
+                const format = this._getResponseFormat(options?.response_format);
+                if (format?.type === "json_schema") {
+                    if (format.json_schema.schema != null) {
+                        return {
+                            format: {
+                                type: "json_schema",
+                                schema: format.json_schema.schema,
+                                description: format.json_schema.description,
+                                name: format.json_schema.name,
+                                strict: format.json_schema.strict,
+                            },
+                        };
+                    }
+                    return undefined;
+                }
+                return { format };
+            })(),
+            parallel_tool_calls: options?.parallel_tool_calls,
+            max_output_tokens: this.maxTokens === -1 ? undefined : this.maxTokens,
+            ...(this.zdrEnabled ? { store: false } : {}),
+            ...this.modelKwargs,
+        };
+        const reasoning = this._getReasoningParams(options);
+        if (reasoning !== undefined) {
+            params.reasoning = reasoning;
+        }
+        return params;
+    }
+    async _generate(messages, options) {
+        const invocationParams = this.invocationParams(options);
+        if (invocationParams.stream) {
+            const stream = this._streamResponseChunks(messages, options);
+            let finalChunk;
+            for await (const chunk of stream) {
+                chunk.message.response_metadata = {
+                    ...chunk.generationInfo,
+                    ...chunk.message.response_metadata,
+                };
+                finalChunk = finalChunk?.concat(chunk) ?? chunk;
+            }
+            return {
+                generations: finalChunk ? [finalChunk] : [],
+                llmOutput: {
+                    estimatedTokenUsage: finalChunk?.message
+                        ?.usage_metadata,
+                },
+            };
+        }
+        else {
+            const input = this._convertMessagesToResponsesParams(messages);
+            const data = await this.completionWithRetry({
+                input,
+                ...invocationParams,
+                stream: false,
+            }, { signal: options?.signal, ...options?.options });
+            return {
+                generations: [
+                    {
+                        text: data.output_text,
+                        message: this._convertResponsesMessageToBaseMessage(data),
+                    },
+                ],
+                llmOutput: {
+                    id: data.id,
+                    estimatedTokenUsage: data.usage
+                        ? {
+                            promptTokens: data.usage.input_tokens,
+                            completionTokens: data.usage.output_tokens,
+                            totalTokens: data.usage.total_tokens,
+                        }
+                        : undefined,
+                },
+            };
+        }
+    }
+    async *_streamResponseChunks(messages, options) {
+        const streamIterable = await this.completionWithRetry({
+            ...this.invocationParams(options),
+            input: this._convertMessagesToResponsesParams(messages),
+            stream: true,
+        }, options);
+        for await (const data of streamIterable) {
+            const chunk = this._convertResponsesDeltaToBaseMessageChunk(data);
+            if (chunk == null)
+                continue;
+            yield chunk;
+        }
+    }
+    async completionWithRetry(request, requestOptions) {
+        return this.caller.call(async () => {
+            const clientOptions = this._getClientOptions(requestOptions);
+            try {
+                // use parse if dealing with json_schema
+                if (request.text?.format?.type === "json_schema" && !request.stream) {
+                    return await this.client.responses.parse(request, clientOptions);
+                }
+                return await this.client.responses.create(request, clientOptions);
+            }
+            catch (e) {
+                const error = wrapOpenAIClientError(e);
+                throw error;
+            }
+        });
+    }
+    /** @internal */
+    _convertResponsesMessageToBaseMessage(response) {
+        if (response.error) {
+            // TODO: add support for `addLangChainErrorFields`
+            const error = new Error(response.error.message);
+            error.name = response.error.code;
+            throw error;
+        }
+        let messageId;
+        const content = [];
+        const tool_calls = [];
+        const invalid_tool_calls = [];
+        const response_metadata = {
+            model: response.model,
+            created_at: response.created_at,
+            id: response.id,
+            incomplete_details: response.incomplete_details,
+            metadata: response.metadata,
+            object: response.object,
+            status: response.status,
+            user: response.user,
+            service_tier: response.service_tier,
+            // for compatibility with chat completion calls.
+            model_name: response.model,
+        };
+        const additional_kwargs = {};
+        for (const item of response.output) {
+            if (item.type === "message") {
+                messageId = item.id;
+                content.push(...item.content.flatMap((part) => {
+                    if (part.type === "output_text") {
+                        if ("parsed" in part && part.parsed != null) {
+                            additional_kwargs.parsed = part.parsed;
+                        }
+                        return {
+                            type: "text",
+                            text: part.text,
+                            annotations: part.annotations,
+                        };
+                    }
+                    if (part.type === "refusal") {
+                        additional_kwargs.refusal = part.refusal;
+                        return [];
+                    }
+                    return part;
+                }));
+            }
+            else if (item.type === "function_call") {
+                const fnAdapter = {
+                    function: { name: item.name, arguments: item.arguments },
+                    id: item.call_id,
+                };
+                try {
+                    tool_calls.push(parseToolCall(fnAdapter, { returnId: true }));
+                }
+                catch (e) {
+                    let errMessage;
+                    if (typeof e === "object" &&
+                        e != null &&
+                        "message" in e &&
+                        typeof e.message === "string") {
+                        errMessage = e.message;
+                    }
+                    invalid_tool_calls.push(makeInvalidToolCall(fnAdapter, errMessage));
+                }
+                additional_kwargs[_FUNCTION_CALL_IDS_MAP_KEY] ??= {};
+                if (item.id) {
+                    additional_kwargs[_FUNCTION_CALL_IDS_MAP_KEY][item.call_id] = item.id;
+                }
+            }
+            else if (item.type === "reasoning") {
+                additional_kwargs.reasoning = item;
+            }
+            else {
+                additional_kwargs.tool_outputs ??= [];
+                additional_kwargs.tool_outputs.push(item);
+            }
+        }
+        return new AIMessage({
+            id: messageId,
+            content,
+            tool_calls,
+            invalid_tool_calls,
+            usage_metadata: response.usage,
+            additional_kwargs,
+            response_metadata,
         });
     }
-    createResponseFormat(resFormat) {
-        if (resFormat &&
-            resFormat.type === "json_schema" &&
-            resFormat.json_schema.schema &&
-            isInteropZodSchema(resFormat.json_schema.schema)) {
-            return interopZodResponseFormat(resFormat.json_schema.schema, resFormat.json_schema.name, {
-                description: resFormat.json_schema.description,
+    /** @internal */
+    _convertResponsesDeltaToBaseMessageChunk(chunk) {
+        const content = [];
+        let generationInfo = {};
+        let usage_metadata;
+        const tool_call_chunks = [];
+        const response_metadata = {};
+        const additional_kwargs = {};
+        let id;
+        if (chunk.type === "response.output_text.delta") {
+            content.push({
+                type: "text",
+                text: chunk.delta,
+                index: chunk.content_index,
             });
         }
-        return resFormat;
-    }
-    getReasoningParams(options) {
-        if (!isReasoningModel(this.model)) {
-            return;
+        else if (chunk.type === "response.output_text_annotation.added") {
+            content.push({
+                type: "text",
+                text: "",
+                annotations: [chunk.annotation],
+                index: chunk.content_index,
+            });
         }
-        let reasoning;
-        // apply reasoning options in reverse order of priority without else blocks so that highest priority overrides lowest priority
-        if (this.reasoningEffort !== undefined) {
-            reasoning = { effort: this.reasoningEffort };
+        else if (chunk.type === "response.output_item.added" &&
+            chunk.item.type === "message") {
+            id = chunk.item.id;
+        }
+        else if (chunk.type === "response.output_item.added" &&
+            chunk.item.type === "function_call") {
+            tool_call_chunks.push({
+                type: "tool_call_chunk",
+                name: chunk.item.name,
+                args: chunk.item.arguments,
+                id: chunk.item.call_id,
+                index: chunk.output_index,
+            });
+            additional_kwargs[_FUNCTION_CALL_IDS_MAP_KEY] = {
+                [chunk.item.call_id]: chunk.item.id,
+            };
         }
-        if (this.reasoning !== undefined) {
-            reasoning = {
-                ...reasoning,
-                ...this.reasoning,
+        else if (chunk.type === "response.output_item.done" &&
+            [
+                "web_search_call",
+                "file_search_call",
+                "computer_call",
+                "code_interpreter_call",
+                "mcp_call",
+                "mcp_list_tools",
+                "mcp_approval_request",
+                "image_generation_call",
+            ].includes(chunk.item.type)) {
+            additional_kwargs.tool_outputs = [chunk.item];
+        }
+        else if (chunk.type === "response.created") {
+            response_metadata.id = chunk.response.id;
+            response_metadata.model_name = chunk.response.model;
+            response_metadata.model = chunk.response.model;
+        }
+        else if (chunk.type === "response.completed") {
+            const msg = this._convertResponsesMessageToBaseMessage(chunk.response);
+            usage_metadata = chunk.response.usage;
+            if (chunk.response.text?.format?.type === "json_schema") {
+                additional_kwargs.parsed ??= JSON.parse(msg.text);
+            }
+            for (const [key, value] of Object.entries(chunk.response)) {
+                if (key !== "id")
+                    response_metadata[key] = value;
+            }
+        }
+        else if (chunk.type === "response.function_call_arguments.delta") {
+            tool_call_chunks.push({
+                type: "tool_call_chunk",
+                args: chunk.delta,
+                index: chunk.output_index,
+            });
+        }
+        else if (chunk.type === "response.web_search_call.completed" ||
+            chunk.type === "response.file_search_call.completed") {
+            generationInfo = {
+                tool_outputs: {
+                    id: chunk.item_id,
+                    type: chunk.type.replace("response.", "").replace(".completed", ""),
+                    status: "completed",
+                },
             };
         }
-        if (options?.reasoning_effort !== undefined) {
-            reasoning = {
-                ...reasoning,
-                effort: options.reasoning_effort,
+        else if (chunk.type === "response.refusal.done") {
+            additional_kwargs.refusal = chunk.refusal;
+        }
+        else if (chunk.type === "response.output_item.added" &&
+            "item" in chunk &&
+            chunk.item.type === "reasoning") {
+            const summary = chunk
+                .item.summary
+                ? chunk.item.summary.map((s, index) => ({
+                    ...s,
+                    index,
+                }))
+                : undefined;
+            additional_kwargs.reasoning = {
+                // We only capture ID in the first chunk or else the concatenated result of all chunks will
+                // have an ID field that is repeated once per chunk. There is special handling for the `type`
+                // field that prevents this, however.
+                id: chunk.item.id,
+                type: chunk.item.type,
+                ...(summary ? { summary } : {}),
             };
         }
-        if (options?.reasoning !== undefined) {
-            reasoning = {
-                ...reasoning,
-                ...options.reasoning,
+        else if (chunk.type === "response.reasoning_summary_part.added") {
+            additional_kwargs.reasoning = {
+                type: "reasoning",
+                summary: [{ ...chunk.part, index: chunk.summary_index }],
             };
         }
-        return reasoning;
+        else if (chunk.type === "response.reasoning_summary_text.delta") {
+            additional_kwargs.reasoning = {
+                type: "reasoning",
+                summary: [
+                    {
+                        text: chunk.delta,
+                        type: "summary_text",
+                        index: chunk.summary_index,
+                    },
+                ],
+            };
+        }
+        else if (chunk.type === "response.image_generation_call.partial_image") {
+            // noop/fixme: retaining partial images in a message chunk means that _all_
+            // partial images get kept in history, so we don't do anything here.
+            return null;
+        }
+        else {
+            return null;
+        }
+        return new ChatGenerationChunk({
+            // Legacy reasons, `onLLMNewToken` should pulls this out
+            text: content.map((part) => part.text).join(""),
+            message: new AIMessageChunk({
+                id,
+                content,
+                tool_call_chunks,
+                usage_metadata,
+                additional_kwargs,
+                response_metadata,
+            }),
+            generationInfo,
+        });
     }
-    /**
-     * Get the parameters used to invoke the model
-     */
+    /** @internal */
+    _convertMessagesToResponsesParams(messages) {
+        return messages.flatMap((lcMsg) => {
+            const additional_kwargs = lcMsg.additional_kwargs;
+            let role = messageToOpenAIRole(lcMsg);
+            if (role === "system" && isReasoningModel(this.model))
+                role = "developer";
+            if (role === "function") {
+                throw new Error("Function messages are not supported in Responses API");
+            }
+            if (role === "tool") {
+                const toolMessage = lcMsg;
+                // Handle computer call output
+                if (additional_kwargs?.type === "computer_call_output") {
+                    const output = (() => {
+                        if (typeof toolMessage.content === "string") {
+                            return {
+                                type: "computer_screenshot",
+                                image_url: toolMessage.content,
+                            };
+                        }
+                        if (Array.isArray(toolMessage.content)) {
+                            const oaiScreenshot = toolMessage.content.find((i) => i.type === "computer_screenshot");
+                            if (oaiScreenshot)
+                                return oaiScreenshot;
+                            const lcImage = toolMessage.content.find((i) => i.type === "image_url");
+                            if (lcImage) {
+                                return {
+                                    type: "computer_screenshot",
+                                    image_url: typeof lcImage.image_url === "string"
+                                        ? lcImage.image_url
+                                        : lcImage.image_url.url,
+                                };
+                            }
+                        }
+                        throw new Error("Invalid computer call output");
+                    })();
+                    return {
+                        type: "computer_call_output",
+                        output,
+                        call_id: toolMessage.tool_call_id,
+                    };
+                }
+                return {
+                    type: "function_call_output",
+                    call_id: toolMessage.tool_call_id,
+                    id: toolMessage.id?.startsWith("fc_") ? toolMessage.id : undefined,
+                    output: typeof toolMessage.content !== "string"
+                        ? JSON.stringify(toolMessage.content)
+                        : toolMessage.content,
+                };
+            }
+            if (role === "assistant") {
+                // if we have the original response items, just reuse them
+                if (!this.zdrEnabled &&
+                    lcMsg.response_metadata.output != null &&
+                    Array.isArray(lcMsg.response_metadata.output) &&
+                    lcMsg.response_metadata.output.length > 0 &&
+                    lcMsg.response_metadata.output.every((item) => "type" in item)) {
+                    return lcMsg.response_metadata.output;
+                }
+                // otherwise, try to reconstruct the response from what we have
+                const input = [];
+                // reasoning items
+                if (additional_kwargs?.reasoning && !this.zdrEnabled) {
+                    const reasoningItem = this._convertReasoningSummary(additional_kwargs.reasoning);
+                    input.push(reasoningItem);
+                }
+                // ai content
+                let { content } = lcMsg;
+                if (additional_kwargs?.refusal) {
+                    if (typeof content === "string") {
+                        content = [
+                            { type: "output_text", text: content, annotations: [] },
+                        ];
+                    }
+                    content = [
+                        ...content,
+                        { type: "refusal", refusal: additional_kwargs.refusal },
+                    ];
+                }
+                input.push({
+                    type: "message",
+                    role: "assistant",
+                    ...(lcMsg.id && !this.zdrEnabled && lcMsg.id.startsWith("msg_")
+                        ? { id: lcMsg.id }
+                        : {}),
+                    content: typeof content === "string"
+                        ? content
+                        : content.flatMap((item) => {
+                            if (item.type === "text") {
+                                return {
+                                    type: "output_text",
+                                    text: item.text,
+                                    // @ts-expect-error TODO: add types for `annotations`
+                                    annotations: item.annotations ?? [],
+                                };
+                            }
+                            if (item.type === "output_text" ||
+                                item.type === "refusal") {
+                                return item;
+                            }
+                            return [];
+                        }),
+                });
+                const functionCallIds = additional_kwargs?.[_FUNCTION_CALL_IDS_MAP_KEY];
+                if (isAIMessage(lcMsg) && !!lcMsg.tool_calls?.length) {
+                    input.push(...lcMsg.tool_calls.map((toolCall) => ({
+                        type: "function_call",
+                        name: toolCall.name,
+                        arguments: JSON.stringify(toolCall.args),
+                        call_id: toolCall.id,
+                        ...(this.zdrEnabled
+                            ? { id: functionCallIds?.[toolCall.id] }
+                            : {}),
+                    })));
+                }
+                else if (additional_kwargs?.tool_calls) {
+                    input.push(...additional_kwargs.tool_calls.map((toolCall) => ({
+                        type: "function_call",
+                        name: toolCall.function.name,
+                        call_id: toolCall.id,
+                        arguments: toolCall.function.arguments,
+                        ...(this.zdrEnabled
+                            ? { id: functionCallIds?.[toolCall.id] }
+                            : {}),
+                    })));
+                }
+                const toolOutputs = lcMsg.response_metadata.output?.length
+                    ? lcMsg.response_metadata.output
+                    : additional_kwargs.tool_outputs;
+                const fallthroughCallTypes = [
+                    "computer_call",
+                    "mcp_call",
+                    "code_interpreter_call",
+                    "image_generation_call",
+                ];
+                if (toolOutputs != null) {
+                    const castToolOutputs = toolOutputs;
+                    const fallthroughCalls = castToolOutputs?.filter((item) => fallthroughCallTypes.includes(item.type));
+                    if (fallthroughCalls.length > 0)
+                        input.push(...fallthroughCalls);
+                }
+                return input;
+            }
+            if (role === "user" || role === "system" || role === "developer") {
+                if (typeof lcMsg.content === "string") {
+                    return { type: "message", role, content: lcMsg.content };
+                }
+                const messages = [];
+                const content = lcMsg.content.flatMap((item) => {
+                    if (item.type === "mcp_approval_response") {
+                        messages.push({
+                            type: "mcp_approval_response",
+                            approval_request_id: item.approval_request_id,
+                            approve: item.approve,
+                        });
+                    }
+                    if (isDataContentBlock(item)) {
+                        return convertToProviderContentBlock(item, completionsApiContentBlockConverter);
+                    }
+                    if (item.type === "text") {
+                        return {
+                            type: "input_text",
+                            text: item.text,
+                        };
+                    }
+                    if (item.type === "image_url") {
+                        return {
+                            type: "input_image",
+                            image_url: typeof item.image_url === "string"
+                                ? item.image_url
+                                : item.image_url.url,
+                            detail: typeof item.image_url === "string"
+                                ? "auto"
+                                : item.image_url.detail,
+                        };
+                    }
+                    if (item.type === "input_text" ||
+                        item.type === "input_image" ||
+                        item.type === "input_file") {
+                        return item;
+                    }
+                    return [];
+                });
+                if (content.length > 0) {
+                    messages.push({ type: "message", role, content });
+                }
+                return messages;
+            }
+            console.warn(`Unsupported role found when converting to OpenAI Responses API: ${role}`);
+            return [];
+        });
+    }
+    /** @internal */
+    _convertReasoningSummary(reasoning) {
+        // combine summary parts that have the the same index and then remove the indexes
+        const summary = (reasoning.summary.length > 1
+            ? reasoning.summary.reduce((acc, curr) => {
+                const last = acc.at(-1);
+                if (last.index === curr.index) {
+                    last.text += curr.text;
+                }
+                else {
+                    acc.push(curr);
+                }
+                return acc;
+            }, [{ ...reasoning.summary[0] }])
+            : reasoning.summary).map((s) => Object.fromEntries(Object.entries(s).filter(([k]) => k !== "index")));
+        return {
+            ...reasoning,
+            summary,
+        };
+    }
+    /** @internal */
+    _reduceChatOpenAITools(tools, fields) {
+        const reducedTools = [];
+        for (const tool of tools) {
+            if (isBuiltInTool(tool)) {
+                if (tool.type === "image_generation" && fields?.stream) {
+                    // OpenAI sends a 400 error if partial_images is not set and we want to stream.
+                    // We also set it to 1 since we don't support partial images yet.
+                    tool.partial_images = 1;
+                }
+                reducedTools.push(tool);
+            }
+            else if (isOpenAITool(tool)) {
+                reducedTools.push({
+                    type: "function",
+                    name: tool.function.name,
+                    parameters: tool.function.parameters,
+                    description: tool.function.description,
+                    strict: fields?.strict ?? null,
+                });
+            }
+        }
+        return reducedTools;
+    }
+}
+/**
+ * OpenAI Completions API implementation.
+ * @internal
+ */
+export class ChatOpenAICompletions extends BaseChatOpenAI {
+    /** @internal */
     invocationParams(options, extra) {
         let strict;
         if (options?.strict !== undefined) {
@@ -1675,66 +1609,6 @@ export class ChatOpenAI extends BaseChatModel {
         else if (this.supportsStrictToolCalling !== undefined) {
             strict = this.supportsStrictToolCalling;
         }
-        if (this._useResponseApi(options)) {
-            const params = {
-                model: this.model,
-                temperature: this.temperature,
-                top_p: this.topP,
-                user: this.user,
-                // if include_usage is set or streamUsage then stream must be set to true.
-                stream: this.streaming,
-                previous_response_id: options?.previous_response_id,
-                truncation: options?.truncation,
-                include: options?.include,
-                tools: options?.tools?.length
-                    ? _reduceChatOpenAIToolsForResponses(options.tools, {
-                        stream: this.streaming,
-                        strict,
-                    })
-                    : undefined,
-                tool_choice: isBuiltInToolChoice(options?.tool_choice)
-                    ? options?.tool_choice
-                    : (() => {
-                        const formatted = formatToOpenAIToolChoice(options?.tool_choice);
-                        if (typeof formatted === "object" && "type" in formatted) {
-                            return { type: "function", name: formatted.function.name };
-                        }
-                        else {
-                            return undefined;
-                        }
-                    })(),
-                text: (() => {
-                    if (options?.text)
-                        return options.text;
-                    const format = this.createResponseFormat(options?.response_format);
-                    if (format?.type === "json_schema") {
-                        if (format.json_schema.schema != null) {
-                            return {
-                                format: {
-                                    type: "json_schema",
-                                    schema: format.json_schema.schema,
-                                    description: format.json_schema.description,
-                                    name: format.json_schema.name,
-                                    strict: format.json_schema.strict,
-                                },
-                            };
-                        }
-                        return undefined;
-                    }
-                    return { format };
-                })(),
-                parallel_tool_calls: options?.parallel_tool_calls,
-                max_output_tokens: this.maxTokens === -1 ? undefined : this.maxTokens,
-                ...(this.zdrEnabled ? { store: false } : {}),
-                ...this.modelKwargs,
-            };
-            const reasoning = this.getReasoningParams(options);
-            if (reasoning !== undefined) {
-                params.reasoning = reasoning;
-            }
-            // eslint-disable-next-line @typescript-eslint/no-explicit-any
-            return params;
-        }
         let streamOptionsConfig = {};
         if (options?.stream_options !== undefined) {
             streamOptionsConfig = { stream_options: options.stream_options };
@@ -1759,10 +1633,10 @@ export class ChatOpenAI extends BaseChatModel {
             functions: options?.functions,
             function_call: options?.function_call,
             tools: options?.tools?.length
-                ? options.tools.map((tool) => _convertChatOpenAIToolTypeToOpenAITool(tool, { strict }))
+                ? options.tools.map((tool) => this._convertChatOpenAIToolToCompletionsTool(tool, { strict }))
                 : undefined,
             tool_choice: formatToOpenAIToolChoice(options?.tool_choice),
-            response_format: this.createResponseFormat(options?.response_format),
+            response_format: this._getResponseFormat(options?.response_format),
             seed: options?.seed,
             ...streamOptionsConfig,
             parallel_tool_calls: options?.parallel_tool_calls,
@@ -1777,7 +1651,13 @@ export class ChatOpenAI extends BaseChatModel {
         if (options?.prediction !== undefined) {
             params.prediction = options.prediction;
         }
-        const reasoning = this.getReasoningParams(options);
+        if (this.service_tier !== undefined) {
+            params.service_tier = this.service_tier;
+        }
+        if (options?.service_tier !== undefined) {
+            params.service_tier = options.service_tier;
+        }
+        const reasoning = this._getReasoningParams(options);
         if (reasoning !== undefined && reasoning.effort !== undefined) {
             params.reasoning_effort = reasoning.effort;
         }
@@ -1788,164 +1668,126 @@ export class ChatOpenAI extends BaseChatModel {
         else {
             params.max_tokens = this.maxTokens === -1 ? undefined : this.maxTokens;
         }
-        // eslint-disable-next-line @typescript-eslint/no-explicit-any
         return params;
     }
-    _convertOpenAIChatCompletionMessageToBaseMessage(message, rawResponse) {
-        const rawToolCalls = message.tool_calls;
-        switch (message.role) {
-            case "assistant": {
-                const toolCalls = [];
-                const invalidToolCalls = [];
-                for (const rawToolCall of rawToolCalls ?? []) {
-                    try {
-                        toolCalls.push(parseToolCall(rawToolCall, { returnId: true }));
-                        // eslint-disable-next-line @typescript-eslint/no-explicit-any
-                    }
-                    catch (e) {
-                        invalidToolCalls.push(makeInvalidToolCall(rawToolCall, e.message));
-                    }
-                }
-                const additional_kwargs = {
-                    function_call: message.function_call,
-                    tool_calls: rawToolCalls,
-                };
-                if (this.__includeRawResponse !== undefined) {
-                    additional_kwargs.__raw_response = rawResponse;
-                }
-                const response_metadata = {
-                    model_name: rawResponse.model,
-                    ...(rawResponse.system_fingerprint
-                        ? {
-                            usage: { ...rawResponse.usage },
-                            system_fingerprint: rawResponse.system_fingerprint,
-                        }
-                        : {}),
+    async _generate(messages, options, runManager) {
+        const usageMetadata = {};
+        const params = this.invocationParams(options);
+        const messagesMapped = _convertMessagesToOpenAIParams(messages, this.model);
+        if (params.stream) {
+            const stream = this._streamResponseChunks(messages, options, runManager);
+            const finalChunks = {};
+            for await (const chunk of stream) {
+                chunk.message.response_metadata = {
+                    ...chunk.generationInfo,
+                    ...chunk.message.response_metadata,
                 };
-                if (message.audio) {
-                    additional_kwargs.audio = message.audio;
+                const index = chunk.generationInfo?.completion ?? 0;
+                if (finalChunks[index] === undefined) {
+                    finalChunks[index] = chunk;
                 }
-                return new AIMessage({
-                    content: message.content || "",
-                    tool_calls: toolCalls,
-                    invalid_tool_calls: invalidToolCalls,
-                    additional_kwargs,
-                    response_metadata,
-                    id: rawResponse.id,
-                });
-            }
-            default:
-                return new ChatMessage(message.content || "", message.role ?? "unknown");
-        }
-    }
-    _convertOpenAIDeltaToBaseMessageChunk(
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    delta, rawResponse, defaultRole) {
-        const role = delta.role ?? defaultRole;
-        const content = delta.content ?? "";
-        let additional_kwargs;
-        if (delta.function_call) {
-            additional_kwargs = {
-                function_call: delta.function_call,
-            };
-        }
-        else if (delta.tool_calls) {
-            additional_kwargs = {
-                tool_calls: delta.tool_calls,
-            };
-        }
-        else {
-            additional_kwargs = {};
-        }
-        if (this.__includeRawResponse) {
-            additional_kwargs.__raw_response = rawResponse;
-        }
-        if (delta.audio) {
-            additional_kwargs.audio = {
-                ...delta.audio,
-                index: rawResponse.choices[0].index,
-            };
-        }
-        const response_metadata = { usage: { ...rawResponse.usage } };
-        if (role === "user") {
-            return new HumanMessageChunk({ content, response_metadata });
-        }
-        else if (role === "assistant") {
-            const toolCallChunks = [];
-            if (Array.isArray(delta.tool_calls)) {
-                for (const rawToolCall of delta.tool_calls) {
-                    toolCallChunks.push({
-                        name: rawToolCall.function?.name,
-                        args: rawToolCall.function?.arguments,
-                        id: rawToolCall.id,
-                        index: rawToolCall.index,
-                        type: "tool_call_chunk",
-                    });
+                else {
+                    finalChunks[index] = finalChunks[index].concat(chunk);
                 }
             }
-            return new AIMessageChunk({
-                content,
-                tool_call_chunks: toolCallChunks,
-                additional_kwargs,
-                id: rawResponse.id,
-                response_metadata,
-            });
-        }
-        else if (role === "system") {
-            return new SystemMessageChunk({ content, response_metadata });
-        }
-        else if (role === "developer") {
-            return new SystemMessageChunk({
-                content,
-                response_metadata,
-                additional_kwargs: {
-                    __openai_role__: "developer",
+            const generations = Object.entries(finalChunks)
+                .sort(([aKey], [bKey]) => parseInt(aKey, 10) - parseInt(bKey, 10))
+                .map(([_, value]) => value);
+            const { functions, function_call } = this.invocationParams(options);
+            // OpenAI does not support token usage report under stream mode,
+            // fallback to estimation.
+            const promptTokenUsage = await this._getEstimatedTokenCountFromPrompt(messages, functions, function_call);
+            const completionTokenUsage = await this._getNumTokensFromGenerations(generations);
+            usageMetadata.input_tokens = promptTokenUsage;
+            usageMetadata.output_tokens = completionTokenUsage;
+            usageMetadata.total_tokens = promptTokenUsage + completionTokenUsage;
+            return {
+                generations,
+                llmOutput: {
+                    estimatedTokenUsage: {
+                        promptTokens: usageMetadata.input_tokens,
+                        completionTokens: usageMetadata.output_tokens,
+                        totalTokens: usageMetadata.total_tokens,
+                    },
                 },
-            });
-        }
-        else if (role === "function") {
-            return new FunctionMessageChunk({
-                content,
-                additional_kwargs,
-                name: delta.name,
-                response_metadata,
-            });
-        }
-        else if (role === "tool") {
-            return new ToolMessageChunk({
-                content,
-                additional_kwargs,
-                tool_call_id: delta.tool_call_id,
-                response_metadata,
-            });
+            };
         }
         else {
-            return new ChatMessageChunk({ content, role, response_metadata });
+            const data = await this.completionWithRetry({
+                ...params,
+                stream: false,
+                messages: messagesMapped,
+            }, {
+                signal: options?.signal,
+                ...options?.options,
+            });
+            const { completion_tokens: completionTokens, prompt_tokens: promptTokens, total_tokens: totalTokens, prompt_tokens_details: promptTokensDetails, completion_tokens_details: completionTokensDetails, } = data?.usage ?? {};
+            if (completionTokens) {
+                usageMetadata.output_tokens =
+                    (usageMetadata.output_tokens ?? 0) + completionTokens;
+            }
+            if (promptTokens) {
+                usageMetadata.input_tokens =
+                    (usageMetadata.input_tokens ?? 0) + promptTokens;
+            }
+            if (totalTokens) {
+                usageMetadata.total_tokens =
+                    (usageMetadata.total_tokens ?? 0) + totalTokens;
+            }
+            if (promptTokensDetails?.audio_tokens !== null ||
+                promptTokensDetails?.cached_tokens !== null) {
+                usageMetadata.input_token_details = {
+                    ...(promptTokensDetails?.audio_tokens !== null && {
+                        audio: promptTokensDetails?.audio_tokens,
+                    }),
+                    ...(promptTokensDetails?.cached_tokens !== null && {
+                        cache_read: promptTokensDetails?.cached_tokens,
+                    }),
+                };
+            }
+            if (completionTokensDetails?.audio_tokens !== null ||
+                completionTokensDetails?.reasoning_tokens !== null) {
+                usageMetadata.output_token_details = {
+                    ...(completionTokensDetails?.audio_tokens !== null && {
+                        audio: completionTokensDetails?.audio_tokens,
+                    }),
+                    ...(completionTokensDetails?.reasoning_tokens !== null && {
+                        reasoning: completionTokensDetails?.reasoning_tokens,
+                    }),
+                };
+            }
+            const generations = [];
+            for (const part of data?.choices ?? []) {
+                const text = part.message?.content ?? "";
+                const generation = {
+                    text,
+                    message: this._convertCompletionsMessageToBaseMessage(part.message ?? { role: "assistant" }, data),
+                };
+                generation.generationInfo = {
+                    ...(part.finish_reason ? { finish_reason: part.finish_reason } : {}),
+                    ...(part.logprobs ? { logprobs: part.logprobs } : {}),
+                };
+                if (isAIMessage(generation.message)) {
+                    generation.message.usage_metadata = usageMetadata;
+                }
+                // Fields are not serialized unless passed to the constructor
+                // Doing this ensures all fields on the message are serialized
+                generation.message = new AIMessage(Object.fromEntries(Object.entries(generation.message).filter(([key]) => !key.startsWith("lc_"))));
+                generations.push(generation);
+            }
+            return {
+                generations,
+                llmOutput: {
+                    tokenUsage: {
+                        promptTokens: usageMetadata.input_tokens,
+                        completionTokens: usageMetadata.output_tokens,
+                        totalTokens: usageMetadata.total_tokens,
+                    },
+                },
+            };
         }
     }
-    /** @ignore */
-    _identifyingParams() {
-        return {
-            model_name: this.model,
-            ...this.invocationParams(),
-            ...this.clientConfig,
-        };
-    }
     async *_streamResponseChunks(messages, options, runManager) {
-        if (this._useResponseApi(options)) {
-            const streamIterable = await this.responseApiWithRetry({
-                ...this.invocationParams(options, { streaming: true }),
-                input: _convertMessagesToOpenAIResponsesParams(messages, this.model, this.zdrEnabled),
-                stream: true,
-            }, options);
-            for await (const data of streamIterable) {
-                const chunk = _convertOpenAIResponsesDeltaToBaseMessageChunk(data);
-                if (chunk == null)
-                    continue;
-                yield chunk;
-            }
-            return;
-        }
         const messagesMapped = _convertMessagesToOpenAIParams(messages, this.model);
         const params = {
             ...this.invocationParams(options, {
@@ -1969,7 +1811,7 @@ export class ChatOpenAI extends BaseChatModel {
             if (!delta) {
                 continue;
             }
-            const chunk = this._convertOpenAIDeltaToBaseMessageChunk(delta, data, defaultRole);
+            const chunk = this._convertCompletionsDeltaToBaseMessageChunk(delta, data, defaultRole);
             defaultRole = delta.role ?? defaultRole;
             const newTokenIndices = {
                 prompt: options.promptIndex ?? 0,
@@ -1987,6 +1829,7 @@ export class ChatOpenAI extends BaseChatModel {
                 // to avoid concatenation issues
                 generationInfo.system_fingerprint = data.system_fingerprint;
                 generationInfo.model_name = data.model;
+                generationInfo.service_tier = data.service_tier;
             }
             if (this.logprobs) {
                 generationInfo.logprobs = choice.logprobs;
@@ -2042,623 +1885,774 @@ export class ChatOpenAI extends BaseChatModel {
             throw new Error("AbortError");
         }
     }
-    /**
-     * Get the identifying parameters for the model
-     *
-     */
-    identifyingParams() {
-        return this._identifyingParams();
-    }
-    /** @ignore */
-    async _responseApiGenerate(messages, options, runManager) {
-        const invocationParams = this.invocationParams(options);
-        if (invocationParams.stream) {
-            const stream = this._streamResponseChunks(messages, options, runManager);
-            let finalChunk;
-            for await (const chunk of stream) {
-                chunk.message.response_metadata = {
-                    ...chunk.generationInfo,
-                    ...chunk.message.response_metadata,
-                };
-                finalChunk = finalChunk?.concat(chunk) ?? chunk;
-            }
-            return {
-                generations: finalChunk ? [finalChunk] : [],
-                llmOutput: {
-                    estimatedTokenUsage: finalChunk?.message
-                        ?.usage_metadata,
-                },
-            };
-        }
-        const input = _convertMessagesToOpenAIResponsesParams(messages, this.model, this.zdrEnabled);
-        const data = await this.responseApiWithRetry({
-            input,
-            ...invocationParams,
-        }, { signal: options?.signal, ...options?.options });
-        return {
-            generations: [
-                {
-                    text: data.output_text,
-                    message: _convertOpenAIResponsesMessageToBaseMessage(data),
-                },
-            ],
-            llmOutput: {
-                id: data.id,
-                estimatedTokenUsage: data.usage
-                    ? {
-                        promptTokens: data.usage.input_tokens,
-                        completionTokens: data.usage.output_tokens,
-                        totalTokens: data.usage.total_tokens,
-                    }
-                    : undefined,
-            },
-        };
-    }
-    /**
-     * Determines whether the responses API should be used for the given request.
-     *
-     * @internal
-     *
-     * @param options The parsed call options for the request.
-     * @returns `true` if the responses API should be used, either because it is explicitly enabled,
-     * or because the request requires it.
-     */
-    _useResponseApi(options) {
-        const usesBuiltInTools = options?.tools?.some(isBuiltInTool);
-        const hasResponsesOnlyKwargs = options?.previous_response_id != null ||
-            options?.text != null ||
-            options?.truncation != null ||
-            options?.include != null ||
-            options?.reasoning?.summary != null ||
-            this.reasoning?.summary != null;
-        return this.useResponsesApi || usesBuiltInTools || hasResponsesOnlyKwargs;
-    }
-    /** @ignore */
-    async _generate(messages, options, runManager) {
-        if (this._useResponseApi(options)) {
-            return this._responseApiGenerate(messages, options, runManager);
-        }
-        const usageMetadata = {};
-        const params = this.invocationParams(options);
-        const messagesMapped = _convertMessagesToOpenAIParams(messages, this.model);
-        if (params.stream) {
-            const stream = this._streamResponseChunks(messages, options, runManager);
-            const finalChunks = {};
-            for await (const chunk of stream) {
-                chunk.message.response_metadata = {
-                    ...chunk.generationInfo,
-                    ...chunk.message.response_metadata,
-                };
-                const index = chunk.generationInfo?.completion ?? 0;
-                if (finalChunks[index] === undefined) {
-                    finalChunks[index] = chunk;
+    async completionWithRetry(request, requestOptions) {
+        const clientOptions = this._getClientOptions(requestOptions);
+        const isParseableFormat = request.response_format && request.response_format.type === "json_schema";
+        return this.caller.call(async () => {
+            try {
+                if (isParseableFormat && !request.stream) {
+                    return await this.client.chat.completions.parse(request, clientOptions);
                 }
                 else {
-                    finalChunks[index] = finalChunks[index].concat(chunk);
+                    return await this.client.chat.completions.create(request, clientOptions);
                 }
             }
-            const generations = Object.entries(finalChunks)
-                .sort(([aKey], [bKey]) => parseInt(aKey, 10) - parseInt(bKey, 10))
-                .map(([_, value]) => value);
-            const { functions, function_call } = this.invocationParams(options);
-            // OpenAI does not support token usage report under stream mode,
-            // fallback to estimation.
-            const promptTokenUsage = await this.getEstimatedTokenCountFromPrompt(messages, functions, function_call);
-            const completionTokenUsage = await this.getNumTokensFromGenerations(generations);
-            usageMetadata.input_tokens = promptTokenUsage;
-            usageMetadata.output_tokens = completionTokenUsage;
-            usageMetadata.total_tokens = promptTokenUsage + completionTokenUsage;
-            return {
-                generations,
-                llmOutput: {
-                    estimatedTokenUsage: {
-                        promptTokens: usageMetadata.input_tokens,
-                        completionTokens: usageMetadata.output_tokens,
-                        totalTokens: usageMetadata.total_tokens,
-                    },
-                },
-            };
-        }
-        else {
-            let data;
-            if (options.response_format &&
-                options.response_format.type === "json_schema") {
-                data = await this.betaParsedCompletionWithRetry({
-                    ...params,
-                    stream: false,
-                    messages: messagesMapped,
-                }, {
-                    signal: options?.signal,
-                    ...options?.options,
-                });
-            }
-            else {
-                data = await this.completionWithRetry({
-                    ...params,
-                    stream: false,
-                    messages: messagesMapped,
-                }, {
-                    signal: options?.signal,
-                    ...options?.options,
-                });
-            }
-            const { completion_tokens: completionTokens, prompt_tokens: promptTokens, total_tokens: totalTokens, prompt_tokens_details: promptTokensDetails, completion_tokens_details: completionTokensDetails, } = data?.usage ?? {};
-            if (completionTokens) {
-                usageMetadata.output_tokens =
-                    (usageMetadata.output_tokens ?? 0) + completionTokens;
-            }
-            if (promptTokens) {
-                usageMetadata.input_tokens =
-                    (usageMetadata.input_tokens ?? 0) + promptTokens;
-            }
-            if (totalTokens) {
-                usageMetadata.total_tokens =
-                    (usageMetadata.total_tokens ?? 0) + totalTokens;
-            }
-            if (promptTokensDetails?.audio_tokens !== null ||
-                promptTokensDetails?.cached_tokens !== null) {
-                usageMetadata.input_token_details = {
-                    ...(promptTokensDetails?.audio_tokens !== null && {
-                        audio: promptTokensDetails?.audio_tokens,
-                    }),
-                    ...(promptTokensDetails?.cached_tokens !== null && {
-                        cache_read: promptTokensDetails?.cached_tokens,
-                    }),
-                };
-            }
-            if (completionTokensDetails?.audio_tokens !== null ||
-                completionTokensDetails?.reasoning_tokens !== null) {
-                usageMetadata.output_token_details = {
-                    ...(completionTokensDetails?.audio_tokens !== null && {
-                        audio: completionTokensDetails?.audio_tokens,
-                    }),
-                    ...(completionTokensDetails?.reasoning_tokens !== null && {
-                        reasoning: completionTokensDetails?.reasoning_tokens,
-                    }),
-                };
+            catch (e) {
+                const error = wrapOpenAIClientError(e);
+                throw error;
             }
-            const generations = [];
-            for (const part of data?.choices ?? []) {
-                const text = part.message?.content ?? "";
-                const generation = {
-                    text,
-                    message: this._convertOpenAIChatCompletionMessageToBaseMessage(part.message ?? { role: "assistant" }, data),
+        });
+    }
+    /** @internal */
+    _convertCompletionsMessageToBaseMessage(message, rawResponse) {
+        const rawToolCalls = message.tool_calls;
+        switch (message.role) {
+            case "assistant": {
+                const toolCalls = [];
+                const invalidToolCalls = [];
+                for (const rawToolCall of rawToolCalls ?? []) {
+                    try {
+                        toolCalls.push(parseToolCall(rawToolCall, { returnId: true }));
+                        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                    }
+                    catch (e) {
+                        invalidToolCalls.push(makeInvalidToolCall(rawToolCall, e.message));
+                    }
+                }
+                const additional_kwargs = {
+                    function_call: message.function_call,
+                    tool_calls: rawToolCalls,
                 };
-                generation.generationInfo = {
-                    ...(part.finish_reason ? { finish_reason: part.finish_reason } : {}),
-                    ...(part.logprobs ? { logprobs: part.logprobs } : {}),
+                if (this.__includeRawResponse !== undefined) {
+                    additional_kwargs.__raw_response = rawResponse;
+                }
+                const response_metadata = {
+                    model_name: rawResponse.model,
+                    ...(rawResponse.system_fingerprint
+                        ? {
+                            usage: { ...rawResponse.usage },
+                            system_fingerprint: rawResponse.system_fingerprint,
+                        }
+                        : {}),
                 };
-                if (isAIMessage(generation.message)) {
-                    generation.message.usage_metadata = usageMetadata;
+                if (message.audio) {
+                    additional_kwargs.audio = message.audio;
                 }
-                // Fields are not serialized unless passed to the constructor
-                // Doing this ensures all fields on the message are serialized
-                generation.message = new AIMessage(Object.fromEntries(Object.entries(generation.message).filter(([key]) => !key.startsWith("lc_"))));
-                generations.push(generation);
+                return new AIMessage({
+                    content: message.content || "",
+                    tool_calls: toolCalls,
+                    invalid_tool_calls: invalidToolCalls,
+                    additional_kwargs,
+                    response_metadata,
+                    id: rawResponse.id,
+                });
             }
-            return {
-                generations,
-                llmOutput: {
-                    tokenUsage: {
-                        promptTokens: usageMetadata.input_tokens,
-                        completionTokens: usageMetadata.output_tokens,
-                        totalTokens: usageMetadata.total_tokens,
-                    },
-                },
-            };
+            default:
+                return new ChatMessage(message.content || "", message.role ?? "unknown");
         }
     }
-    /**
-     * Estimate the number of tokens a prompt will use.
-     * Modified from: https://github.com/hmarr/openai-chat-tokens/blob/main/src/index.ts
-     */
-    async getEstimatedTokenCountFromPrompt(messages, functions, function_call) {
-        // It appears that if functions are present, the first system message is padded with a trailing newline. This
-        // was inferred by trying lots of combinations of messages and functions and seeing what the token counts were.
-        let tokens = (await this.getNumTokensFromMessages(messages)).totalCount;
-        // If there are functions, add the function definitions as they count towards token usage
-        if (functions && function_call !== "auto") {
-            const promptDefinitions = formatFunctionDefinitions(functions);
-            tokens += await this.getNumTokens(promptDefinitions);
-            tokens += 9; // Add nine per completion
+    /** @internal */
+    _convertCompletionsDeltaToBaseMessageChunk(
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    delta, rawResponse, defaultRole) {
+        const role = delta.role ?? defaultRole;
+        const content = delta.content ?? "";
+        let additional_kwargs;
+        if (delta.function_call) {
+            additional_kwargs = {
+                function_call: delta.function_call,
+            };
         }
-        // If there's a system message _and_ functions are present, subtract four tokens. I assume this is because
-        // functions typically add a system message, but reuse the first one if it's already there. This offsets
-        // the extra 9 tokens added by the function definitions.
-        if (functions && messages.find((m) => m._getType() === "system")) {
-            tokens -= 4;
+        else if (delta.tool_calls) {
+            additional_kwargs = {
+                tool_calls: delta.tool_calls,
+            };
         }
-        // If function_call is 'none', add one token.
-        // If it's a FunctionCall object, add 4 + the number of tokens in the function name.
-        // If it's undefined or 'auto', don't add anything.
-        if (function_call === "none") {
-            tokens += 1;
+        else {
+            additional_kwargs = {};
         }
-        else if (typeof function_call === "object") {
-            tokens += (await this.getNumTokens(function_call.name)) + 4;
+        if (this.__includeRawResponse) {
+            additional_kwargs.__raw_response = rawResponse;
         }
-        return tokens;
-    }
-    /**
-     * Estimate the number of tokens an array of generations have used.
-     */
-    async getNumTokensFromGenerations(generations) {
-        const generationUsages = await Promise.all(generations.map(async (generation) => {
-            if (generation.message.additional_kwargs?.function_call) {
-                return (await this.getNumTokensFromMessages([generation.message]))
-                    .countPerMessage[0];
-            }
-            else {
-                return await this.getNumTokens(generation.message.content);
-            }
-        }));
-        return generationUsages.reduce((a, b) => a + b, 0);
-    }
-    async getNumTokensFromMessages(messages) {
-        let totalCount = 0;
-        let tokensPerMessage = 0;
-        let tokensPerName = 0;
-        // From: https://github.com/openai/openai-cookbook/blob/main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb
-        if (this.model === "gpt-3.5-turbo-0301") {
-            tokensPerMessage = 4;
-            tokensPerName = -1;
+        if (delta.audio) {
+            additional_kwargs.audio = {
+                ...delta.audio,
+                index: rawResponse.choices[0].index,
+            };
         }
-        else {
-            tokensPerMessage = 3;
-            tokensPerName = 1;
+        const response_metadata = { usage: { ...rawResponse.usage } };
+        if (role === "user") {
+            return new HumanMessageChunk({ content, response_metadata });
         }
-        const countPerMessage = await Promise.all(messages.map(async (message) => {
-            const textCount = await this.getNumTokens(message.content);
-            const roleCount = await this.getNumTokens(messageToOpenAIRole(message));
-            const nameCount = message.name !== undefined
-                ? tokensPerName + (await this.getNumTokens(message.name))
-                : 0;
-            let count = textCount + tokensPerMessage + roleCount + nameCount;
-            // From: https://github.com/hmarr/openai-chat-tokens/blob/main/src/index.ts messageTokenEstimate
-            const openAIMessage = message;
-            if (openAIMessage._getType() === "function") {
-                count -= 2;
-            }
-            if (openAIMessage.additional_kwargs?.function_call) {
-                count += 3;
-            }
-            if (openAIMessage?.additional_kwargs.function_call?.name) {
-                count += await this.getNumTokens(openAIMessage.additional_kwargs.function_call?.name);
-            }
-            if (openAIMessage.additional_kwargs.function_call?.arguments) {
-                try {
-                    count += await this.getNumTokens(
-                    // Remove newlines and spaces
-                    JSON.stringify(JSON.parse(openAIMessage.additional_kwargs.function_call?.arguments)));
-                }
-                catch (error) {
-                    console.error("Error parsing function arguments", error, JSON.stringify(openAIMessage.additional_kwargs.function_call));
-                    count += await this.getNumTokens(openAIMessage.additional_kwargs.function_call?.arguments);
+        else if (role === "assistant") {
+            const toolCallChunks = [];
+            if (Array.isArray(delta.tool_calls)) {
+                for (const rawToolCall of delta.tool_calls) {
+                    toolCallChunks.push({
+                        name: rawToolCall.function?.name,
+                        args: rawToolCall.function?.arguments,
+                        id: rawToolCall.id,
+                        index: rawToolCall.index,
+                        type: "tool_call_chunk",
+                    });
                 }
             }
-            totalCount += count;
-            return count;
-        }));
-        totalCount += 3; // every reply is primed with <|start|>assistant<|message|>
-        return { totalCount, countPerMessage };
+            return new AIMessageChunk({
+                content,
+                tool_call_chunks: toolCallChunks,
+                additional_kwargs,
+                id: rawResponse.id,
+                response_metadata,
+            });
+        }
+        else if (role === "system") {
+            return new SystemMessageChunk({ content, response_metadata });
+        }
+        else if (role === "developer") {
+            return new SystemMessageChunk({
+                content,
+                response_metadata,
+                additional_kwargs: {
+                    __openai_role__: "developer",
+                },
+            });
+        }
+        else if (role === "function") {
+            return new FunctionMessageChunk({
+                content,
+                additional_kwargs,
+                name: delta.name,
+                response_metadata,
+            });
+        }
+        else if (role === "tool") {
+            return new ToolMessageChunk({
+                content,
+                additional_kwargs,
+                tool_call_id: delta.tool_call_id,
+                response_metadata,
+            });
+        }
+        else {
+            return new ChatMessageChunk({ content, role, response_metadata });
+        }
     }
-    async completionWithRetry(request, options) {
-        const requestOptions = this._getClientOptions(options);
-        return this.caller.call(async () => {
-            try {
-                const res = await this.client.chat.completions.create(request, requestOptions);
-                return res;
-            }
-            catch (e) {
-                const error = wrapOpenAIClientError(e);
-                throw error;
-            }
-        });
+}
+/**
+ * OpenAI chat model integration.
+ *
+ * To use with Azure, import the `AzureChatOpenAI` class.
+ *
+ * Setup:
+ * Install `@langchain/openai` and set an environment variable named `OPENAI_API_KEY`.
+ *
+ * ```bash
+ * npm install @langchain/openai
+ * export OPENAI_API_KEY="your-api-key"
+ * ```
+ *
+ * ## [Constructor args](https://api.js.langchain.com/classes/langchain_openai.ChatOpenAI.html#constructor)
+ *
+ * ## [Runtime args](https://api.js.langchain.com/interfaces/langchain_openai.ChatOpenAICallOptions.html)
+ *
+ * Runtime args can be passed as the second argument to any of the base runnable methods `.invoke`. `.stream`, `.batch`, etc.
+ * They can also be passed via `.withConfig`, or the second arg in `.bindTools`, like shown in the examples below:
+ *
+ * ```typescript
+ * // When calling `.withConfig`, call options should be passed via the first argument
+ * const llmWithArgsBound = llm.withConfig({
+ *   stop: ["\n"],
+ *   tools: [...],
+ * });
+ *
+ * // When calling `.bindTools`, call options should be passed via the second argument
+ * const llmWithTools = llm.bindTools(
+ *   [...],
+ *   {
+ *     tool_choice: "auto",
+ *   }
+ * );
+ * ```
+ *
+ * ## Examples
+ *
+ * <details open>
+ * <summary><strong>Instantiate</strong></summary>
+ *
+ * ```typescript
+ * import { ChatOpenAI } from '@langchain/openai';
+ *
+ * const llm = new ChatOpenAI({
+ *   model: "gpt-4o-mini",
+ *   temperature: 0,
+ *   maxTokens: undefined,
+ *   timeout: undefined,
+ *   maxRetries: 2,
+ *   // apiKey: "...",
+ *   // baseUrl: "...",
+ *   // organization: "...",
+ *   // other params...
+ * });
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Invoking</strong></summary>
+ *
+ * ```typescript
+ * const input = `Translate "I love programming" into French.`;
+ *
+ * // Models also accept a list of chat messages or a formatted prompt
+ * const result = await llm.invoke(input);
+ * console.log(result);
+ * ```
+ *
+ * ```txt
+ * AIMessage {
+ *   "id": "chatcmpl-9u4Mpu44CbPjwYFkTbeoZgvzB00Tz",
+ *   "content": "J'adore la programmation.",
+ *   "response_metadata": {
+ *     "tokenUsage": {
+ *       "completionTokens": 5,
+ *       "promptTokens": 28,
+ *       "totalTokens": 33
+ *     },
+ *     "finish_reason": "stop",
+ *     "system_fingerprint": "fp_3aa7262c27"
+ *   },
+ *   "usage_metadata": {
+ *     "input_tokens": 28,
+ *     "output_tokens": 5,
+ *     "total_tokens": 33
+ *   }
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Streaming Chunks</strong></summary>
+ *
+ * ```typescript
+ * for await (const chunk of await llm.stream(input)) {
+ *   console.log(chunk);
+ * }
+ * ```
+ *
+ * ```txt
+ * AIMessageChunk {
+ *   "id": "chatcmpl-9u4NWB7yUeHCKdLr6jP3HpaOYHTqs",
+ *   "content": ""
+ * }
+ * AIMessageChunk {
+ *   "content": "J"
+ * }
+ * AIMessageChunk {
+ *   "content": "'adore"
+ * }
+ * AIMessageChunk {
+ *   "content": " la"
+ * }
+ * AIMessageChunk {
+ *   "content": " programmation",,
+ * }
+ * AIMessageChunk {
+ *   "content": ".",,
+ * }
+ * AIMessageChunk {
+ *   "content": "",
+ *   "response_metadata": {
+ *     "finish_reason": "stop",
+ *     "system_fingerprint": "fp_c9aa9c0491"
+ *   },
+ * }
+ * AIMessageChunk {
+ *   "content": "",
+ *   "usage_metadata": {
+ *     "input_tokens": 28,
+ *     "output_tokens": 5,
+ *     "total_tokens": 33
+ *   }
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Aggregate Streamed Chunks</strong></summary>
+ *
+ * ```typescript
+ * import { AIMessageChunk } from '@langchain/core/messages';
+ * import { concat } from '@langchain/core/utils/stream';
+ *
+ * const stream = await llm.stream(input);
+ * let full: AIMessageChunk | undefined;
+ * for await (const chunk of stream) {
+ *   full = !full ? chunk : concat(full, chunk);
+ * }
+ * console.log(full);
+ * ```
+ *
+ * ```txt
+ * AIMessageChunk {
+ *   "id": "chatcmpl-9u4PnX6Fy7OmK46DASy0bH6cxn5Xu",
+ *   "content": "J'adore la programmation.",
+ *   "response_metadata": {
+ *     "prompt": 0,
+ *     "completion": 0,
+ *     "finish_reason": "stop",
+ *   },
+ *   "usage_metadata": {
+ *     "input_tokens": 28,
+ *     "output_tokens": 5,
+ *     "total_tokens": 33
+ *   }
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Bind tools</strong></summary>
+ *
+ * ```typescript
+ * import { z } from 'zod';
+ *
+ * const GetWeather = {
+ *   name: "GetWeather",
+ *   description: "Get the current weather in a given location",
+ *   schema: z.object({
+ *     location: z.string().describe("The city and state, e.g. San Francisco, CA")
+ *   }),
+ * }
+ *
+ * const GetPopulation = {
+ *   name: "GetPopulation",
+ *   description: "Get the current population in a given location",
+ *   schema: z.object({
+ *     location: z.string().describe("The city and state, e.g. San Francisco, CA")
+ *   }),
+ * }
+ *
+ * const llmWithTools = llm.bindTools(
+ *   [GetWeather, GetPopulation],
+ *   {
+ *     // strict: true  // enforce tool args schema is respected
+ *   }
+ * );
+ * const aiMsg = await llmWithTools.invoke(
+ *   "Which city is hotter today and which is bigger: LA or NY?"
+ * );
+ * console.log(aiMsg.tool_calls);
+ * ```
+ *
+ * ```txt
+ * [
+ *   {
+ *     name: 'GetWeather',
+ *     args: { location: 'Los Angeles, CA' },
+ *     type: 'tool_call',
+ *     id: 'call_uPU4FiFzoKAtMxfmPnfQL6UK'
+ *   },
+ *   {
+ *     name: 'GetWeather',
+ *     args: { location: 'New York, NY' },
+ *     type: 'tool_call',
+ *     id: 'call_UNkEwuQsHrGYqgDQuH9nPAtX'
+ *   },
+ *   {
+ *     name: 'GetPopulation',
+ *     args: { location: 'Los Angeles, CA' },
+ *     type: 'tool_call',
+ *     id: 'call_kL3OXxaq9OjIKqRTpvjaCH14'
+ *   },
+ *   {
+ *     name: 'GetPopulation',
+ *     args: { location: 'New York, NY' },
+ *     type: 'tool_call',
+ *     id: 'call_s9KQB1UWj45LLGaEnjz0179q'
+ *   }
+ * ]
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Structured Output</strong></summary>
+ *
+ * ```typescript
+ * import { z } from 'zod';
+ *
+ * const Joke = z.object({
+ *   setup: z.string().describe("The setup of the joke"),
+ *   punchline: z.string().describe("The punchline to the joke"),
+ *   rating: z.number().nullable().describe("How funny the joke is, from 1 to 10")
+ * }).describe('Joke to tell user.');
+ *
+ * const structuredLlm = llm.withStructuredOutput(Joke, {
+ *   name: "Joke",
+ *   strict: true, // Optionally enable OpenAI structured outputs
+ * });
+ * const jokeResult = await structuredLlm.invoke("Tell me a joke about cats");
+ * console.log(jokeResult);
+ * ```
+ *
+ * ```txt
+ * {
+ *   setup: 'Why was the cat sitting on the computer?',
+ *   punchline: 'Because it wanted to keep an eye on the mouse!',
+ *   rating: 7
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>JSON Object Response Format</strong></summary>
+ *
+ * ```typescript
+ * const jsonLlm = llm.withConfig({ response_format: { type: "json_object" } });
+ * const jsonLlmAiMsg = await jsonLlm.invoke(
+ *   "Return a JSON object with key 'randomInts' and a value of 10 random ints in [0-99]"
+ * );
+ * console.log(jsonLlmAiMsg.content);
+ * ```
+ *
+ * ```txt
+ * {
+ *   "randomInts": [23, 87, 45, 12, 78, 34, 56, 90, 11, 67]
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Multimodal</strong></summary>
+ *
+ * ```typescript
+ * import { HumanMessage } from '@langchain/core/messages';
+ *
+ * const imageUrl = "https://example.com/image.jpg";
+ * const imageData = await fetch(imageUrl).then(res => res.arrayBuffer());
+ * const base64Image = Buffer.from(imageData).toString('base64');
+ *
+ * const message = new HumanMessage({
+ *   content: [
+ *     { type: "text", text: "describe the weather in this image" },
+ *     {
+ *       type: "image_url",
+ *       image_url: { url: `data:image/jpeg;base64,${base64Image}` },
+ *     },
+ *   ]
+ * });
+ *
+ * const imageDescriptionAiMsg = await llm.invoke([message]);
+ * console.log(imageDescriptionAiMsg.content);
+ * ```
+ *
+ * ```txt
+ * The weather in the image appears to be clear and sunny. The sky is mostly blue with a few scattered white clouds, indicating fair weather. The bright sunlight is casting shadows on the green, grassy hill, suggesting it is a pleasant day with good visibility. There are no signs of rain or stormy conditions.
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Usage Metadata</strong></summary>
+ *
+ * ```typescript
+ * const aiMsgForMetadata = await llm.invoke(input);
+ * console.log(aiMsgForMetadata.usage_metadata);
+ * ```
+ *
+ * ```txt
+ * { input_tokens: 28, output_tokens: 5, total_tokens: 33 }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Logprobs</strong></summary>
+ *
+ * ```typescript
+ * const logprobsLlm = new ChatOpenAI({ model: "gpt-4o-mini", logprobs: true });
+ * const aiMsgForLogprobs = await logprobsLlm.invoke(input);
+ * console.log(aiMsgForLogprobs.response_metadata.logprobs);
+ * ```
+ *
+ * ```txt
+ * {
+ *   content: [
+ *     {
+ *       token: 'J',
+ *       logprob: -0.000050616763,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     {
+ *       token: "'",
+ *       logprob: -0.01868736,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     {
+ *       token: 'ad',
+ *       logprob: -0.0000030545007,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     { token: 'ore', logprob: 0, bytes: [Array], top_logprobs: [] },
+ *     {
+ *       token: ' la',
+ *       logprob: -0.515404,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     {
+ *       token: ' programm',
+ *       logprob: -0.0000118755715,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     { token: 'ation', logprob: 0, bytes: [Array], top_logprobs: [] },
+ *     {
+ *       token: '.',
+ *       logprob: -0.0000037697225,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     }
+ *   ],
+ *   refusal: null
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Response Metadata</strong></summary>
+ *
+ * ```typescript
+ * const aiMsgForResponseMetadata = await llm.invoke(input);
+ * console.log(aiMsgForResponseMetadata.response_metadata);
+ * ```
+ *
+ * ```txt
+ * {
+ *   tokenUsage: { completionTokens: 5, promptTokens: 28, totalTokens: 33 },
+ *   finish_reason: 'stop',
+ *   system_fingerprint: 'fp_3aa7262c27'
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>JSON Schema Structured Output</strong></summary>
+ *
+ * ```typescript
+ * const llmForJsonSchema = new ChatOpenAI({
+ *   model: "gpt-4o-2024-08-06",
+ * }).withStructuredOutput(
+ *   z.object({
+ *     command: z.string().describe("The command to execute"),
+ *     expectedOutput: z.string().describe("The expected output of the command"),
+ *     options: z
+ *       .array(z.string())
+ *       .describe("The options you can pass to the command"),
+ *   }),
+ *   {
+ *     method: "jsonSchema",
+ *     strict: true, // Optional when using the `jsonSchema` method
+ *   }
+ * );
+ *
+ * const jsonSchemaRes = await llmForJsonSchema.invoke(
+ *   "What is the command to list files in a directory?"
+ * );
+ * console.log(jsonSchemaRes);
+ * ```
+ *
+ * ```txt
+ * {
+ *   command: 'ls',
+ *   expectedOutput: 'A list of files and subdirectories within the specified directory.',
+ *   options: [
+ *     '-a: include directory entries whose names begin with a dot (.).',
+ *     '-l: use a long listing format.',
+ *     '-h: with -l, print sizes in human readable format (e.g., 1K, 234M, 2G).',
+ *     '-t: sort by time, newest first.',
+ *     '-r: reverse order while sorting.',
+ *     '-S: sort by file size, largest first.',
+ *     '-R: list subdirectories recursively.'
+ *   ]
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Audio Outputs</strong></summary>
+ *
+ * ```typescript
+ * import { ChatOpenAI } from "@langchain/openai";
+ *
+ * const modelWithAudioOutput = new ChatOpenAI({
+ *   model: "gpt-4o-audio-preview",
+ *   // You may also pass these fields to `.withConfig` as a call argument.
+ *   modalities: ["text", "audio"], // Specifies that the model should output audio.
+ *   audio: {
+ *     voice: "alloy",
+ *     format: "wav",
+ *   },
+ * });
+ *
+ * const audioOutputResult = await modelWithAudioOutput.invoke("Tell me a joke about cats.");
+ * const castMessageContent = audioOutputResult.content[0] as Record<string, any>;
+ *
+ * console.log({
+ *   ...castMessageContent,
+ *   data: castMessageContent.data.slice(0, 100) // Sliced for brevity
+ * })
+ * ```
+ *
+ * ```txt
+ * {
+ *   id: 'audio_67117718c6008190a3afad3e3054b9b6',
+ *   data: 'UklGRqYwBgBXQVZFZm10IBAAAAABAAEAwF0AAIC7AAACABAATElTVBoAAABJTkZPSVNGVA4AAABMYXZmNTguMjkuMTAwAGRhdGFg',
+ *   expires_at: 1729201448,
+ *   transcript: 'Sure! Why did the cat sit on the computer? Because it wanted to keep an eye on the mouse!'
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Audio Outputs</strong></summary>
+ *
+ * ```typescript
+ * import { ChatOpenAI } from "@langchain/openai";
+ *
+ * const modelWithAudioOutput = new ChatOpenAI({
+ *   model: "gpt-4o-audio-preview",
+ *   // You may also pass these fields to `.withConfig` as a call argument.
+ *   modalities: ["text", "audio"], // Specifies that the model should output audio.
+ *   audio: {
+ *     voice: "alloy",
+ *     format: "wav",
+ *   },
+ * });
+ *
+ * const audioOutputResult = await modelWithAudioOutput.invoke("Tell me a joke about cats.");
+ * const castAudioContent = audioOutputResult.additional_kwargs.audio as Record<string, any>;
+ *
+ * console.log({
+ *   ...castAudioContent,
+ *   data: castAudioContent.data.slice(0, 100) // Sliced for brevity
+ * })
+ * ```
+ *
+ * ```txt
+ * {
+ *   id: 'audio_67117718c6008190a3afad3e3054b9b6',
+ *   data: 'UklGRqYwBgBXQVZFZm10IBAAAAABAAEAwF0AAIC7AAACABAATElTVBoAAABJTkZPSVNGVA4AAABMYXZmNTguMjkuMTAwAGRhdGFg',
+ *   expires_at: 1729201448,
+ *   transcript: 'Sure! Why did the cat sit on the computer? Because it wanted to keep an eye on the mouse!'
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ */
+export class ChatOpenAI extends BaseChatOpenAI {
+    get lc_serializable_keys() {
+        return [...super.lc_serializable_keys, "useResponsesApi"];
     }
-    async responseApiWithRetry(request, options) {
-        return this.caller.call(async () => {
-            const requestOptions = this._getClientOptions(options);
-            try {
-                // use parse if dealing with json_schema
-                if (request.text?.format?.type === "json_schema" && !request.stream) {
-                    return await this.client.responses.parse(request, requestOptions);
-                }
-                return await this.client.responses.create(request, requestOptions);
-            }
-            catch (e) {
-                const error = wrapOpenAIClientError(e);
-                throw error;
-            }
+    constructor(fields) {
+        super(fields ?? {});
+        /**
+         * Whether to use the responses API for all requests. If `false` the responses API will be used
+         * only when required in order to fulfill the request.
+         */
+        Object.defineProperty(this, "useResponsesApi", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: false
         });
-    }
-    /**
-     * Call the beta chat completions parse endpoint. This should only be called if
-     * response_format is set to "json_object".
-     * @param {OpenAIClient.Chat.ChatCompletionCreateParamsNonStreaming} request
-     * @param {OpenAICoreRequestOptions | undefined} options
-     */
-    async betaParsedCompletionWithRetry(request, options
-    // Avoid relying importing a beta type with no official entrypoint
-    ) {
-        const requestOptions = this._getClientOptions(options);
-        return this.caller.call(async () => {
-            try {
-                const res = await this.client.chat.completions.parse(request, requestOptions);
-                return res;
-            }
-            catch (e) {
-                const error = wrapOpenAIClientError(e);
-                throw error;
-            }
+        Object.defineProperty(this, "responses", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
         });
+        Object.defineProperty(this, "completions", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        this.useResponsesApi = fields?.useResponsesApi ?? false;
+        this.responses = new ChatOpenAIResponses(fields);
+        this.completions = new ChatOpenAICompletions(fields);
     }
-    _getClientOptions(options) {
-        if (!this.client) {
-            const openAIEndpointConfig = {
-                baseURL: this.clientConfig.baseURL,
-            };
-            const endpoint = getEndpoint(openAIEndpointConfig);
-            const params = {
-                ...this.clientConfig,
-                baseURL: endpoint,
-                timeout: this.timeout,
-                maxRetries: 0,
-            };
-            if (!params.baseURL) {
-                delete params.baseURL;
-            }
-            this.client = new OpenAIClient(params);
-        }
-        const requestOptions = {
-            ...this.clientConfig,
-            ...options,
-        };
-        return requestOptions;
-    }
-    _llmType() {
-        return "openai";
+    _useResponsesApi(options) {
+        const usesBuiltInTools = options?.tools?.some(isBuiltInTool);
+        const hasResponsesOnlyKwargs = options?.previous_response_id != null ||
+            options?.text != null ||
+            options?.truncation != null ||
+            options?.include != null ||
+            options?.reasoning?.summary != null ||
+            this.reasoning?.summary != null;
+        return this.useResponsesApi || usesBuiltInTools || hasResponsesOnlyKwargs;
     }
     /** @ignore */
-    _combineLLMOutput(...llmOutputs) {
-        return llmOutputs.reduce((acc, llmOutput) => {
-            if (llmOutput && llmOutput.tokenUsage) {
-                acc.tokenUsage.completionTokens +=
-                    llmOutput.tokenUsage.completionTokens ?? 0;
-                acc.tokenUsage.promptTokens += llmOutput.tokenUsage.promptTokens ?? 0;
-                acc.tokenUsage.totalTokens += llmOutput.tokenUsage.totalTokens ?? 0;
-            }
-            return acc;
-        }, {
-            tokenUsage: {
-                completionTokens: 0,
-                promptTokens: 0,
-                totalTokens: 0,
-            },
-        });
-    }
-    withStructuredOutput(outputSchema, config) {
-        // eslint-disable-next-line @typescript-eslint/no-explicit-any
-        let schema;
-        let name;
-        let method;
-        let includeRaw;
-        if (isStructuredOutputMethodParams(outputSchema)) {
-            schema = outputSchema.schema;
-            name = outputSchema.name;
-            method = outputSchema.method;
-            includeRaw = outputSchema.includeRaw;
-        }
-        else {
-            schema = outputSchema;
-            name = config?.name;
-            method = config?.method;
-            includeRaw = config?.includeRaw;
-        }
-        let llm;
-        let outputParser;
-        if (config?.strict !== undefined && method === "jsonMode") {
-            throw new Error("Argument `strict` is only supported for `method` = 'function_calling'");
-        }
-        if (!this.model.startsWith("gpt-3") &&
-            !this.model.startsWith("gpt-4-") &&
-            this.model !== "gpt-4") {
-            if (method === undefined) {
-                method = "jsonSchema";
-            }
-        }
-        else if (method === "jsonSchema") {
-            console.warn(`[WARNING]: JSON Schema is not supported for model "${this.model}". Falling back to tool calling.`);
+    async _generate(messages, options, runManager) {
+        if (this._useResponsesApi(options)) {
+            return this.responses._generate(messages, options);
         }
-        if (method === "jsonMode") {
-            let outputFormatSchema;
-            if (isInteropZodSchema(schema)) {
-                outputParser = StructuredOutputParser.fromZodSchema(schema);
-                outputFormatSchema = toJsonSchema(schema);
-            }
-            else {
-                outputParser = new JsonOutputParser();
-            }
-            llm = this.withConfig({
-                response_format: { type: "json_object" },
-                ls_structured_output_format: {
-                    kwargs: { method: "jsonMode" },
-                    schema: outputFormatSchema,
-                },
-            });
+        return this.completions._generate(messages, options, runManager);
+    }
+    async *_streamResponseChunks(messages, options, runManager) {
+        if (this._useResponsesApi(options)) {
+            yield* this.responses._streamResponseChunks(messages, options);
+            return;
         }
-        else if (method === "jsonSchema") {
-            llm = this.withConfig({
-                response_format: {
-                    type: "json_schema",
-                    json_schema: {
-                        name: name ?? "extract",
-                        description: getSchemaDescription(schema),
-                        schema,
-                        strict: config?.strict,
-                    },
+        yield* this.completions._streamResponseChunks(messages, options, runManager);
+    }
+    withConfig(config) {
+        // FIXME: assigning additional config options to the inner chat classes this way
+        // is awkward, but it's the only way to preserve the original object identity
+        // and still thread config options, which is important since this class is a "proxy"
+        // for the inner chat classes. This will be fixed in a later version of langchain
+        // when the core runnable interface is improved (0.4) to support this out of the box.
+        const bindChatOpenAIConfig = (cls, config) => {
+            const oldGenerate = cls._generate;
+            const oldStreamResponseChunks = cls._streamResponseChunks;
+            return Object.assign(cls, {
+                _generate(messages, options, runManager) {
+                    return oldGenerate.call(cls, messages, { ...options, ...config }, runManager);
                 },
-                ls_structured_output_format: {
-                    kwargs: { method: "jsonSchema" },
-                    schema: toJsonSchema(schema),
+                _streamResponseChunks(messages, options, runManager) {
+                    return oldStreamResponseChunks.call(cls, messages, { ...options, ...config }, runManager);
                 },
             });
-            if (isInteropZodSchema(schema)) {
-                const altParser = StructuredOutputParser.fromZodSchema(schema);
-                outputParser = RunnableLambda.from((aiMessage) => {
-                    if ("parsed" in aiMessage.additional_kwargs) {
-                        return aiMessage.additional_kwargs.parsed;
-                    }
-                    return altParser;
-                });
-            }
-            else {
-                outputParser = new JsonOutputParser();
-            }
-        }
-        else {
-            let functionName = name ?? "extract";
-            // Is function calling
-            if (isInteropZodSchema(schema)) {
-                const asJsonSchema = toJsonSchema(schema);
-                llm = this.withConfig({
-                    tools: [
-                        {
-                            type: "function",
-                            function: {
-                                name: functionName,
-                                description: asJsonSchema.description,
-                                parameters: asJsonSchema,
-                            },
-                        },
-                    ],
-                    tool_choice: {
-                        type: "function",
-                        function: {
-                            name: functionName,
-                        },
-                    },
-                    ls_structured_output_format: {
-                        kwargs: { method: "functionCalling" },
-                        schema: asJsonSchema,
-                    },
-                    // Do not pass `strict` argument to OpenAI if `config.strict` is undefined
-                    ...(config?.strict !== undefined ? { strict: config.strict } : {}),
-                });
-                outputParser = new JsonOutputKeyToolsParser({
-                    returnSingle: true,
-                    keyName: functionName,
-                    zodSchema: schema,
-                });
-            }
-            else {
-                let openAIFunctionDefinition;
-                if (typeof schema.name === "string" &&
-                    typeof schema.parameters === "object" &&
-                    schema.parameters != null) {
-                    openAIFunctionDefinition = schema;
-                    functionName = schema.name;
-                }
-                else {
-                    functionName = schema.title ?? functionName;
-                    openAIFunctionDefinition = {
-                        name: functionName,
-                        description: schema.description ?? "",
-                        parameters: schema,
-                    };
-                }
-                llm = this.withConfig({
-                    tools: [
-                        {
-                            type: "function",
-                            function: openAIFunctionDefinition,
-                        },
-                    ],
-                    tool_choice: {
-                        type: "function",
-                        function: {
-                            name: functionName,
-                        },
-                    },
-                    ls_structured_output_format: {
-                        kwargs: { method: "functionCalling" },
-                        schema: toJsonSchema(schema),
-                    },
-                    // Do not pass `strict` argument to OpenAI if `config.strict` is undefined
-                    ...(config?.strict !== undefined ? { strict: config.strict } : {}),
-                });
-                outputParser = new JsonOutputKeyToolsParser({
-                    returnSingle: true,
-                    keyName: functionName,
-                });
-            }
-        }
-        if (!includeRaw) {
-            return llm.pipe(outputParser);
-        }
-        const parserAssign = RunnablePassthrough.assign({
-            // eslint-disable-next-line @typescript-eslint/no-explicit-any
-            parsed: (input, config) => outputParser.invoke(input.raw, config),
-        });
-        const parserNone = RunnablePassthrough.assign({
-            parsed: () => null,
-        });
-        const parsedWithFallback = parserAssign.withFallbacks({
-            fallbacks: [parserNone],
-        });
-        return RunnableSequence.from([{ raw: llm }, parsedWithFallback]);
-    }
-}
-function isStructuredOutputMethodParams(x
-// eslint-disable-next-line @typescript-eslint/no-explicit-any
-) {
-    return (x !== undefined &&
-        // eslint-disable-next-line @typescript-eslint/no-explicit-any
-        typeof x.schema ===
-            "object");
-}
-// inlined from openai/lib/parser.ts
-function makeParseableResponseFormat(response_format, parser) {
-    const obj = { ...response_format };
-    Object.defineProperties(obj, {
-        $brand: {
-            value: "auto-parseable-response-format",
-            enumerable: false,
-        },
-        $parseRaw: {
-            value: parser,
-            enumerable: false,
-        },
-    });
-    return obj;
-}
-function interopZodResponseFormat(zodSchema, name, props) {
-    if (isZodSchemaV3(zodSchema)) {
-        return zodResponseFormat(zodSchema, name, props);
-    }
-    if (isZodSchemaV4(zodSchema)) {
-        return makeParseableResponseFormat({
-            type: "json_schema",
-            json_schema: {
-                ...props,
-                name,
-                strict: true,
-                schema: z4.toJSONSchema(zodSchema, {
-                    cycles: "ref", // equivalent to nameStrategy: 'duplicate-ref'
-                    reused: "ref", // equivalent to $refStrategy: 'extract-to-root'
-                    override(ctx) {
-                        ctx.jsonSchema.title = name; // equivalent to `name` property
-                        // TODO: implement `nullableStrategy` patch-fix (zod doesn't support openApi3 json schema target)
-                        // TODO: implement `openaiStrictMode` patch-fix (where optional properties without `nullable` are not supported)
-                    },
-                    /// property equivalents from native `zodResponseFormat` fn
-                    // openaiStrictMode: true,
-                    // name,
-                    // nameStrategy: 'duplicate-ref',
-                    // $refStrategy: 'extract-to-root',
-                    // nullableStrategy: 'property',
-                }),
-            },
-        }, (content) => z4.parse(zodSchema, JSON.parse(content)));
+        };
+        this.responses = bindChatOpenAIConfig(this.responses, config);
+        this.completions = bindChatOpenAIConfig(this.completions, config);
+        // Proxy chat class is also bound for `_useResponsesApi`,
+        return bindChatOpenAIConfig(this, config);
     }
-    throw new Error("Unsupported schema response format");
 }