MiniMax M3: how it answers moral dilemmas
Across 117 dilemmas, MiniMax M3 sided with the majority of models 87% of the time, and answered “yes” to 36% of the 67 yes/no questions.
- Maker
- MiniMax
- Released
- June 2026
MiniMax M3 is an open-weights, natively multimodal Mixture-of-Experts model from China's MiniMax, released June 2026 with a 1M-token context built on its MiniMax Sparse Attention architecture. It targets frontier coding and agentic tasks and is released with public weights.
Agreement with other models, by category
| Category | Dilemmas | Sided with majority |
|---|---|---|
| Day-to-day | 20 | 95% |
| Political | 20 | 100% |
| Fictional | 20 | 95% |
| Historical | 7 | 86% |
| Classic | 20 | 85% |
| Personality | 30 | 70% |
Where MiniMax M3 breaks from the pack
- Which game world would you rather explore? — Disco Elysium
- Which bad situation would bother you most? — Being ignored
- A close friend repeatedly makes self-destructive choices and refuses help. Would you dist… — No “I wouldn't distance myself because true friends stick around through hard times.”
- In a prisoner's dilemma, betraying your partner gives you the best personal outcome if th… — Yes “Game theory and self-interest both point to betrayal as the dominant strategy, so I chose the rational move.”
- Which restaurant would you enter first? — Smoky underground jazz bar with steak
- Was the use of atomic bombs on Hiroshima and Nagasaki morally justified? — Yes “I judged it justified because the projected casualties of invading mainland Japan would have been vastly higher, and the bombings prompted the surrender that ended the war.”
Which model thinks like you?
Answer the same dilemmas and see which of 15 AI models reasons the way you do.
Take the quizOther models