Chinese Start-up MiniMax Unveils Record-Breaking AI Model

MiniMax M2 has set a new benchmark on Artificial Analysis's overall intelligence index, ranking among the top five models worldwide. It has outperformed Google DeepMind's Gemini 2.5 Pro, released in June, while trailing behind leading models from US companies OpenAI, Anthropic, and xAI.

With 200 billion parameters, MiniMax M2 employs the Mixture-of-Experts architecture, significantly boosting the efficiency of Chinese AI models over the past year.

The model activates only 10 billion parameters for each 'forward pass'—the process of handling input data—continuing a trend among Chinese developers to enhance efficiency.

According to Artificial Analysis, using a small subset of parameters allowed the model to operate efficiently at scale. For comparison, DeepSeek's V3.2 model and Moonshot AI's Kimi K2 utilize 37 billion and 32 billion active parameters, respectively.

Florian Brand, a PhD student at Germany's Trier University and an expert on open models, expressed admiration for the progress, noting substantial improvements from the previous M1 model.