Open Source

Minimax M3 surprises by showing no political censorship in benchmark test

A Chinese LLM breaks the mold—M3 appears uncensored where others are not.

Deep Dive

A user working on a Chinese/CCP AI bias benchmark reports that one model stands out as an outlier while all other Minimax models are censored, as is typical for Chinese LLMs.

Key Points
  • Minimax M3 is the only model from the company that appears uncensored in a Chinese political bias benchmark.
  • Other Minimax models follow standard censorship patterns typical of Chinese LLMs.
  • The finding suggests either a deliberate policy exception or a testing/ deployment error by Minimax.

Why It Matters

If real, M3 offers an unprecedented look at uncensored Chinese AI—possible policy shift or oversight.