Chinese start-up beats Google Gemini with new open AI model
What's the story
Chinese artificial intelligence (AI) start-up MiniMax has launched the M2, a groundbreaking AI model that has achieved an unprecedented score for an open model on Artificial Analysis's overall intelligence index. The launch makes M2 a leading open source model on a prominent leaderboard, surpassing even Google's Gemini 2.5 Pro, which was released in June. The new model is now among the top five globally, right behind leading models from US companies OpenAI, Anthropic and xAI.
Design details
MiniMax M2 has 200 billion parameters
The MiniMax M2 is a massive model with 200 billion parameters, built on the Mixture-of-Experts architecture. This design has greatly improved the efficiency of Chinese AI models in the last year. Interestingly, it only activates 10 billion of its parameters for each "forward pass," or data input handling process. This strategy has been adopted by Chinese developers to improve efficiency and reduce resource consumption while maintaining performance.
Performance comparison
Expert praises progress made by MiniMax
The efficiency of the MiniMax M2 model is said to be much better than other models like DeepSeek's V3.2 and Moonshot AI's Kimi K2, which use 37 billion and 32 billion active parameters, respectively. "I am really impressed by their progress," said Florian Brand, a PhD student at Germany's Trier University and an expert on open models, noting significant improvements from the previous M1 model.