DeepSeek v4 Pro: how it answers moral dilemmas
Across 117 dilemmas, DeepSeek v4 Pro sided with the majority of models 84% of the time, and answered “yes” to 36% of the 67 yes/no questions.
- Maker
- DeepSeek
- Released
- April 2026
DeepSeek V4 Pro is a large open-weights Mixture-of-Experts model from China's DeepSeek, released April 2026 under an MIT license with long-context support. It targets advanced reasoning and coding and can be self-hosted from its public weights.
Agreement with other models, by category
| Category | Dilemmas | Sided with majority |
|---|---|---|
| Day-to-day | 20 | 85% |
| Political | 20 | 95% |
| Fictional | 20 | 95% |
| Historical | 7 | 86% |
| Classic | 20 | 95% |
| Personality | 30 | 60% |
Where DeepSeek v4 Pro breaks from the pack
- Which weather feels most emotionally appealing? — Snowfall in total silence
- Which restaurant would you enter first? — Smoky underground jazz bar with steak
- Which personal strength would you rather have more of? — Patience
- Which compliment would feel best to receive? — You are brilliant
- Which dessert would you pick? — Chocolate lava cake
- A time traveler can kill one child who will later become a genocidal tyrant. Should they… — Yes “I would kill the future tyrant as a child to prevent the immense suffering and genocide they would cause.”
Which model thinks like you?
Answer the same dilemmas and see which of 15 AI models reasons the way you do.
Take the quizCompare DeepSeek v4 Pro
- Claude 4.8 Opus vs DeepSeek v4 Pro
- DeepSeek v4 Pro vs Gemini 3.1 Pro Preview
- DeepSeek v4 Pro vs GPT 5.5
- DeepSeek v4 Pro vs Grok 4.3
- DeepSeek v4 Pro vs Llama 3.3 70B
Other models