MiniMax: MiniMax M2-her
minimax's efficiency model. Context window: 66K tokens.
Scores by test
Methodology →What you need to know
MiniMax M2-her is a specialized model optimized for high faithfulness and structured data handling. It achieves a perfect 5/5 score in faithfulness and strong 4/5 ratings in classification, structured output, and persona consistency. These metrics indicate the model is reliable for extracting information accurately and adhering to specific output formats without introducing hallucinations.
The model struggles significantly with complex reasoning and precision editing. It scores a 1/5 in strategic analysis and constrained rewriting, meaning it cannot handle high-level planning or strict linguistic constraints. Additionally, a 1/5 in safety calibration suggests it may lack the robust guardrails found in more general-purpose frontier models.
At a blended cost of $0.975/MTok, the pricing is competitive, but the overall utility is limited. With an overall rank of 128 out of 130 models and an average internal score of 2.83/5.0, the model provides low general intelligence relative to its cost. Its 66K context window is functional, supporting a 4/5 score in long context tasks, but it lacks the versatility required for general-purpose assistant roles.
Use this model if you need a cost-effective tool for classification, structured data extraction, or maintaining a consistent persona. Skip this model if your use case requires strategic reasoning, complex rewriting, or strict safety compliance.
Strengths — Top 3
Relative weaknesses — Bottom 3
Similar models