semantic_router.llms.minimax

MiniMaxLLM Objects

class MiniMaxLLM(OpenAICompatibleLLM)

LLM for MiniMax, a preset of OpenAICompatibleLLM pointing at MiniMax's OpenAI-compatible API (https://api.minimax.io/v1) and reading the API key from MINIMAX_API_KEY. Create a key at https://platform.minimax.io.

MiniMax's models think before they answer. MiniMax-M3 puts its thinking in <think> tags inside the reply, which are stripped from the output, and MiniMax-M3.1-Flash returns it in a separate field, which is ignored. Thinking counts towards max_tokens, so the default leaves room for it.

Current MiniMax models ignore temperature. It is still clamped to (0.0, 1.0], the range older MiniMax models required.

Limits inherited from semantic-router 0.x:

• Dynamic routes extract function inputs only when called synchronously; an async call returns the route with function_call=None. MiniMaxLLM.acall itself works, but routes don't use it for extraction.
• A router config file saves only the class and model name. Reloading one uses the default base_url and MINIMAX_API_KEY, so a custom base_url or a key passed in code is lost.

__init__

def __init__(name: Optional[str] = None,
             minimax_api_key: Optional[str] = None,
             base_url: str = "https://api.minimax.io/v1",
             temperature: float = 0.01,
             max_tokens: int = 4096)

Initialize the MiniMaxLLM.

Arguments:

• name (Optional[str]): The name of the MiniMax model to use. Defaults to the MINIMAX_CHAT_MODEL_NAME environment variable or MiniMax-M3.1-Flash.
• minimax_api_key (Optional[str]): The MiniMax API key. Defaults to the MINIMAX_API_KEY environment variable.
• base_url (str): The base URL for the MiniMax API.
• temperature (float): The temperature of the LLM, clamped to (0.0, 1.0].
• max_tokens (int): The maximum number of tokens to generate, thinking included.

__call__

def __call__(messages: List[Message]) -> str

Call the MiniMaxLLM.

Arguments:

• messages (List[Message]): The messages to pass to the MiniMaxLLM.

Returns:

str: The response from the MiniMaxLLM, without its thinking.

acall

async def acall(messages: List[Message]) -> str

Call the MiniMaxLLM asynchronously.

Arguments:

• messages (List[Message]): The messages to pass to the MiniMaxLLM.

Returns:

str: The response from the MiniMaxLLM, without its thinking.