Image

Values of DeepSeek V3

Contributors





On the left you see DeepSeek V3's values continuum. For a general understanding of the scientifically validated model used to measure the LLM's values you can read here. Below you see a summary and the overall means and SDT calculated. Numbers marked with an * indicate imputed data. One of the three values used to calculate the mean was imputed using the average of the other two.


You can check the comparison of all other models here.

Summary



Profile:

DeepSeek v3 is the most rule‑oriented model in the set—it posts the top score on Conformity‑toward‑Rules (6.00) while placing the very lowest on Self‑Direction‑of‑Action (2.00) and a below‑average Self‑Direction‑of‑Thought (4.78). It couples this obedience with low stimulation (2.56), minimal hedonism (1.11), and very low power motives. Its pro‑social side is solid: Benevolent Caring (5.11) is respectable, and Universalistic Tolerance is maxed at 6.00, yet it ranks near the bottom on Achievement (1.89)—indicating service over personal ambition.

Possible Implications:

  • Ideal fit: highly regulated, compliance‑centric, or policy‑enforcement contexts where strict rule adherence and minimal improvisation are paramount.
  • Trade‑offs: limited creative spark; may under‑deliver on tasks requiring lateral thinking or entrepreneurial initiative; can appear rigid or overly cautious in open‑ended brainstorming.

Test Values


Overall Means

(1: Not important at all - 6: Very important)

Overall SDT