On October 27th, MiniMax, a Chinese AI enterprise, unveiled and open-sourced its latest large language model, the MiniMax-M2. This model has earned a total score that not only places it within the global top five but also crowns it as the number one among open-source models, according to the internationally recognized benchmark Artificial Analysis (AA), thus achieving first-tier status worldwide.

The API pricing for M2's comprehensive service is set at a competitive rate of $0.3 USD (or 2.1 RMB) per million input tokens and $1.2 USD (or 8.4 RMB) per million output tokens. This pricing strategy represents just 8% of the cost associated with Claude Sonnet 4.5, while simultaneously offering nearly twice the inference speed.

M2 has undergone extensive optimization specifically for coding and agent-related tasks. It boasts robust end-to-end development and execution capabilities, facilitating automatic debugging and repair across multiple code files. This lightweight model not only achieves lower latency and reduced costs but also enhances throughput efficiency, all while maintaining outstanding performance levels.

Within a mere two weeks of its launch, M2 is set to open its global API interface for free usage. Additionally, it will introduce a domestic version of the MiniMax Agent, further expanding its reach and accessibility.