Implementation:Langchain ai Langchain BaseOpenAI LLM

Knowledge Sources	Langchain_ai_Langchain
Domains	LLM, OpenAI
Last Updated	2026-02-11 00:00 GMT

Overview

BaseOpenAI and OpenAI are LangChain LLM classes that provide text completion via the OpenAI Completions API.

Description

The base.py file in the langchain-openai partner package defines two classes: BaseOpenAI (the base implementation) and OpenAI (the concrete, serializable subclass). BaseOpenAI extends BaseLLM from langchain-core and wraps the openai.OpenAI and openai.AsyncOpenAI clients to provide text completions (non-chat) using models like gpt-3.5-turbo-instruct. The class supports batch processing of prompts, streaming output, synchronous and asynchronous generation, token counting via tiktoken, configurable sampling parameters (temperature, top_p, frequency_penalty, presence_penalty, logit_bias, seed, logprobs), and automatic prompt batching based on configurable batch sizes. It also provides static model-to-context-size mapping and dynamic max token calculation based on prompt length.

Usage

Import OpenAI when you need text completions (non-chat) from OpenAI models. Use BaseOpenAI as a base class for custom OpenAI-compatible LLM implementations.

Code Reference

Source Location

• Repository: Langchain_ai_Langchain
• File: libs/partners/openai/langchain_openai/llms/base.py
• Lines: 1-872

Signature

class BaseOpenAI(BaseLLM):
    client: Any = Field(default=None, exclude=True)
    async_client: Any = Field(default=None, exclude=True)
    model_name: str = Field(default="gpt-3.5-turbo-instruct", alias="model")
    temperature: float = 0.7
    max_tokens: int = 256
    top_p: float = 1
    frequency_penalty: float = 0
    presence_penalty: float = 0
    n: int = 1
    best_of: int = 1
    model_kwargs: dict[str, Any] = Field(default_factory=dict)
    openai_api_key: SecretStr | None | Callable[[], str] = Field(alias="api_key", ...)
    openai_api_base: str | None = Field(alias="base_url", ...)
    openai_organization: str | None = Field(alias="organization", ...)
    openai_proxy: str | None = Field(...)
    batch_size: int = 20
    request_timeout: float | tuple[float, float] | Any | None = Field(default=None, alias="timeout")
    logit_bias: dict[str, float] | None = None
    max_retries: int = 2
    seed: int | None = None
    logprobs: int | None = None
    streaming: bool = False
    allowed_special: Literal["all"] | set[str] = set()
    disallowed_special: Literal["all"] | Collection[str] = "all"
    tiktoken_model_name: str | None = None
    default_headers: Mapping[str, str] | None = None
    default_query: Mapping[str, object] | None = None
    http_client: Any | None = None
    http_async_client: Any | None = None
    extra_body: Mapping[str, Any] | None = None


class OpenAI(BaseOpenAI):
    ...

Import

from langchain_openai import OpenAI
# or for the base class:
from langchain_openai.llms.base import BaseOpenAI

I/O Contract

Inputs

Name	Type	Required	Description
model_name	str	No	OpenAI model name. Defaults to "gpt-3.5-turbo-instruct". Alias: model.
openai_api_key	SecretStr or Callable or None	No	API key. Inferred from OPENAI_API_KEY env var. Alias: api_key.
openai_api_base	str or None	No	Base URL for API requests. Alias: base_url.
openai_organization	str or None	No	Organization ID. Inferred from OPENAI_ORG_ID env var. Alias: organization.
temperature	float	No	Sampling temperature. Defaults to 0.7.
max_tokens	int	No	Maximum tokens to generate. Defaults to 256. -1 for max possible.
top_p	float	No	Nucleus sampling parameter. Defaults to 1.
frequency_penalty	float	No	Frequency penalty for repeated tokens. Defaults to 0.
presence_penalty	float	No	Presence penalty for repeated tokens. Defaults to 0.
n	int	No	Number of completions to generate per prompt. Defaults to 1.
best_of	int	No	Server-side generation count, returns best. Defaults to 1.
batch_size	int	No	Batch size for multiple prompts. Defaults to 20.
seed	int or None	No	Seed for deterministic generation.
logprobs	int or None	No	Number of most likely tokens to return log probabilities for.
streaming	bool	No	Whether to stream results. Defaults to False.
max_retries	int	No	Maximum request retries. Defaults to 2.
request_timeout	float or tuple or None	No	Request timeout. Alias: timeout.

Outputs

Name	Type	Description
LLMResult	LLMResult	Contains lists of Generation objects with generated text, finish reason, logprobs, token usage, and model name.
GenerationChunk	Iterator[GenerationChunk]	When streaming, yields chunks of generated text with finish reason and logprobs.

Key Methods

get_token_ids

Uses tiktoken to tokenize text for the specified model, supporting custom allowed/disallowed special tokens.

modelname_to_contextsize

Static method mapping OpenAI model names to their maximum context window sizes. Supports fine-tuned model name parsing.

max_tokens_for_prompt

Calculates the maximum number of tokens that can be generated for a given prompt based on the model's context size.

get_sub_prompts

Splits a list of prompts into batches of size batch_size and handles the max_tokens=-1 case.

Usage Examples

Basic Usage

from langchain_openai import OpenAI

model = OpenAI(
    model="gpt-3.5-turbo-instruct",
    temperature=0,
    max_retries=2,
)

response = model.invoke("The meaning of life is ")
print(response)

Streaming

from langchain_openai import OpenAI

model = OpenAI(model="gpt-3.5-turbo-instruct", streaming=True)

for chunk in model.stream("The meaning of life is "):
    print(chunk, end="")

Async Usage

from langchain_openai import OpenAI

model = OpenAI(model="gpt-3.5-turbo-instruct")

response = await model.ainvoke("The meaning of life is ")
print(response)

Related Pages

• Environment:Langchain_ai_Langchain_OpenAI_API_Credentials