MiniMax: MiniMax M2-her

MiniMax M2-her is a text model from MiniMax, listed on api14 under the model id minimax/minimax-m2-her. It is a dialogue-first large language model, which means its design centres on conversation rather than on general purpose text transformation. MiniMax positions it for immersive roleplay, character-driven chat, and expressive multi-turn conversations, with an emphasis on staying consistent in tone and personality across a long exchange. If your work involves a persistent persona or a chat experience where voice matters, this is the kind of model the provider built it for.

On the capability side, the model accepts text and returns text, so it is a pure language model with no image or audio handling. It supports a context window of 65,536 tokens, which is enough to carry a substantial conversation history, a detailed character brief, or a long back and forth without trimming on every turn. The maximum output is 2,048 tokens per response. That ceiling suits chat replies, in-character messages, and scene continuations, though it is not intended for generating very long documents in a single call.

The practical uses follow directly from that profile. Character chatbots and roleplay companions are the obvious fit, since the model is tuned to hold a persona and keep its tone steady over many turns. Interactive storytelling, visual novel style dialogue, and text adventure games also map well onto a model that handles multi-turn conversation and expressive phrasing. Beyond entertainment, the same strengths apply to support agents or tutoring assistants that need a defined voice and a consistent manner across a session. Anywhere the quality of the dialogue itself is the product, this model is a reasonable candidate.

You call it through api14's OpenAI-compatible API using the model id minimax/minimax-m2-her, so existing OpenAI SDKs and clients work with only the base URL and key changed. The request parameters it supports are max_tokens, temperature, and top_p. Use max_tokens to cap reply length within the 2,048 token output limit, temperature to move between restrained and more varied phrasing, and top_p for nucleus sampling as an alternative or complement to temperature. Prompts are plain text messages, and multi-turn history is passed the usual way.

One limit worth noting is that the model is text only, so tasks involving images, audio, or other input types need a different model. The 2,048 token output cap also makes it a poor fit for long form generation such as full articles or reports in one pass. If your workload is document summarisation, structured extraction, or code generation rather than conversation, a general purpose text model may serve you better. For dialogue-centred work, M2-her is the more targeted choice.