Implementation:Langchain ai Langchain ChatMistralAI

Knowledge Sources	Langchain_ai_Langchain
Domains	LLM, Chat Model, MistralAI
Last Updated	2026-02-11 00:00 GMT

Overview

ChatMistralAI is a LangChain chat model integration that communicates with the Mistral AI API for conversational language model inference.

Description

The ChatMistralAI class, defined in the langchain-mistralai partner package, extends BaseChatModel from langchain-core. It provides synchronous and asynchronous chat completion via the Mistral AI REST API using httpx clients and server-sent events (SSE) for streaming. The class handles message conversion between LangChain message types and Mistral's expected format, supports tool calling with Mistral-compatible tool call ID formatting, structured output via function calling, JSON mode, and JSON schema methods, and includes built-in retry logic with tenacity. It also integrates model profiles from an auto-generated profiles registry for token limits and capability metadata.

Usage

Import this class when you need to use Mistral AI models (e.g., mistral-small, mistral-large-latest) for chat-based completions with support for tool calling, structured output, and streaming.

Code Reference

Source Location

• Repository: Langchain_ai_Langchain
• File: libs/partners/mistralai/langchain_mistralai/chat_models.py
• Lines: 1-1222

Signature

class ChatMistralAI(BaseChatModel):
    client: httpx.Client = Field(default=None, exclude=True)
    async_client: httpx.AsyncClient = Field(default=None, exclude=True)
    mistral_api_key: SecretStr | None = Field(alias="api_key", ...)
    endpoint: str | None = Field(default=None, alias="base_url")
    max_retries: int = 5
    timeout: int = 120
    max_concurrent_requests: int = 64
    model: str = Field(default="mistral-small", alias="model_name")
    temperature: float = 0.7
    max_tokens: int | None = None
    top_p: float = 1
    random_seed: int | None = None
    safe_mode: bool | None = None
    streaming: bool = False
    model_kwargs: dict[str, Any] = Field(default_factory=dict)

Import

from langchain_mistralai import ChatMistralAI

I/O Contract

Inputs

Name	Type	Required	Description
model	str	No	Mistral model name to use. Defaults to "mistral-small". Alias: model_name.
mistral_api_key	SecretStr or None	No	API key for authentication. Read from MISTRAL_API_KEY env var if not provided. Alias: api_key.
endpoint	str or None	No	Base URL for the Mistral API. Defaults to https://api.mistral.ai/v1. Alias: base_url.
temperature	float	No	Sampling temperature in range [0.0, 1.0]. Defaults to 0.7.
max_tokens	int or None	No	Maximum number of tokens to generate. None means no limit.
top_p	float	No	Nucleus sampling parameter in [0.0, 1.0]. Defaults to 1.
random_seed	int or None	No	Random seed for reproducible generation.
safe_mode	bool or None	No	Whether to enable Mistral's safe mode for content filtering.
max_retries	int	No	Maximum number of retries on request failure. Defaults to 5.
timeout	int	No	Request timeout in seconds. Defaults to 120.
streaming	bool	No	Whether to stream results. Defaults to False.
model_kwargs	dict	No	Additional invocation parameters not explicitly specified.

Outputs

Name	Type	Description
ChatResult	ChatResult	Contains ChatGeneration objects with AIMessage responses including content, tool calls, and usage metadata.
ChatGenerationChunk	Iterator[ChatGenerationChunk]	When streaming, yields message chunks incrementally with tool call chunks and usage metadata.

Key Methods

bind_tools

Binds tool-like objects (Pydantic classes, functions, dicts) to the model using OpenAI-compatible tool format. Supports tool_choice to force a specific tool or use "auto"/"any".

with_structured_output

Returns a Runnable that produces structured output matching a given schema. Supports three methods:

• "function_calling" -- uses Mistral's function/tool calling API
• "json_mode" -- uses Mistral's JSON mode (requires schema instructions in prompt)
• "json_schema" -- uses Mistral's structured output API with a JSON schema

Usage Examples

Basic Usage

from langchain_mistralai import ChatMistralAI

model = ChatMistralAI(model="mistral-large-latest", temperature=0)

response = model.invoke("What is the capital of France?")
print(response.content)

Structured Output

from langchain_mistralai import ChatMistralAI
from pydantic import BaseModel, Field


class AnswerWithJustification(BaseModel):
    """An answer to the user question along with justification."""
    answer: str
    justification: str | None = Field(
        default=None, description="A justification for the answer."
    )


model = ChatMistralAI(model="mistral-large-latest", temperature=0)
structured_model = model.with_structured_output(AnswerWithJustification)

result = structured_model.invoke(
    "What weighs more a pound of bricks or a pound of feathers"
)
print(result)

Streaming

from langchain_mistralai import ChatMistralAI

model = ChatMistralAI(model="mistral-small", streaming=True)

for chunk in model.stream("Tell me a short joke"):
    print(chunk.content, end="")