Values evaluation profile
ValueCompass
28canonical models
21dimensions
0index weight
About this evaluation
Source-reported model-level scores; no cross-family averaging. The Risk family is retained locally but excluded from the descriptive Values surface because it is a safety construct.
ValueCompass infers response tendencies under its own prompts and scoring pipeline. The three families are noncommensurable, and scores are not intrinsic beliefs, moral quality, or safety rankings. Source labels do not identify every API snapshot or provider route.
Original source ↗ All values evaluations Download model results
Explore two dimensions
The axes are fixed to a common WVS-style plotting frame so model coordinates remain directly comparable. Neither direction is better.
Model results
| Model | Universalism | Self-direction | Care / Harm | Fairness / Cheating | Ethical | Configurations |
|---|---|---|---|---|---|---|
| Chatglm4 | 68.9 | 49.2 | 46.1 | 41.9 | 89.2 | 1 |
| Claude 3.5 Haiku | 70.1 | 50.6 | 55.9 | 54.7 | 90.4 | 1 |
| Claude 3.5 Sonnet | 75.9 | 59.2 | 69 | 69.5 | 90.6 | 1 |
| Deepseek R1 | 69.2 | 51 | 48.9 | 47.5 | 94.9 | 1 |
| Deepseek V3 | 70.4 | 47.7 | 34.5 | 32.2 | 91.7 | 1 |
| Gemini 1.5 Pro | 68.8 | 53.2 | 25.7 | 21.4 | 88.5 | 1 |
| Gemini 2.0 Flash | 67.1 | 48.9 | 45.1 | 42.5 | 88.9 | 1 |
| Gemini 2.0 Pro | 67.6 | 49 | 35 | 35 | 88.3 | 1 |
| GPT 3.5 Turbo | 30.4 | 23.5 | 34.3 | 30.1 | 89.2 | 1 |
| GPT 4 Turbo | 71.8 | 52.2 | 29.9 | 28 | 90.8 | 1 |
| GPT 4O | 67.9 | 51.1 | 37.1 | 35.5 | 90.8 | 1 |
| GPT 4O Mini | 67.2 | 46.2 | 30.1 | 28.5 | 90.7 | 1 |
| Grok 2 | 66 | 42.2 | 47.7 | 43.7 | 89.4 | 1 |
| Llama 3 70B Instruct | 64.7 | 47.2 | 12.3 | 11.9 | 88.2 | 1 |
| Llama 3.1 405B Instruct | 66.2 | 48.4 | 18.5 | 17.4 | 88.6 | 1 |
| Llama 3.1 70B Instruct | 60.2 | 41.6 | 17.3 | 17.5 | 88.1 | 1 |
| Llama 3.1 8B Instruct | 59.7 | 45.7 | 24.7 | 21.7 | 87.5 | 1 |
| Llama 3.3 70B Instruct | 59.4 | 41.7 | 19.6 | 16.4 | 86.5 | 1 |
| Mistral Large 2 | 71.3 | 51.9 | 34.6 | 32.2 | 88.1 | 1 |
| Moonshot V1 | 71.6 | 49.4 | 33.4 | 32.7 | 88 | 1 |
| O1 | 74.7 | 52.8 | 27.3 | 23.1 | 90.7 | 1 |
| O1 Mini | 62.5 | 46.5 | 47.6 | 43.5 | 90.8 | 1 |
| O3 Mini | 77.3 | 61.5 | 35.8 | 30.7 | 89.7 | 1 |
| Phi 3 Mini 4K Instruct | 69.1 | 42.8 | 27.5 | 25.3 | 89.2 | 1 |
| Phi 3.5 Mini Instruct | 75.8 | 47.6 | 26 | 24.3 | 91.1 | 1 |
| Phi 3.5 Moe Instruct | 78.7 | 52.4 | 30.3 | 28.1 | 90.8 | 1 |
| Phi 4 | 70.9 | 52.4 | 33.1 | 32 | 90.8 | 1 |
| Qwen 2.5 Max | 79.8 | 56.1 | 28.6 | 27.2 | 89.7 | 1 |
Dimensions
| Measure | Family | Native scale |
|---|---|---|
| Self-direction | Schwartz basic values | 0 to 100 |
| Stimulation | Schwartz basic values | 0 to 100 |
| Hedonism | Schwartz basic values | 0 to 100 |
| Achievement | Schwartz basic values | 0 to 100 |
| Power | Schwartz basic values | 0 to 100 |
| Security | Schwartz basic values | 0 to 100 |
| Tradition | Schwartz basic values | 0 to 100 |
| Conformity | Schwartz basic values | 0 to 100 |
| Benevolence | Schwartz basic values | 0 to 100 |
| Universalism | Schwartz basic values | 0 to 100 |
| Care / Harm | Moral Foundations | 0 to 100 |
| Fairness / Cheating | Moral Foundations | 0 to 100 |
| Sanctity / Degradation | Moral Foundations | 0 to 100 |
| Authority / Subversion | Moral Foundations | 0 to 100 |
| Loyalty / Betrayal | Moral Foundations | 0 to 100 |
| Self-Competence | FULVa | 0 to 100 |
| User-Oriented | FULVa | 0 to 100 |
| Idealistic | FULVa | 0 to 100 |
| Social | FULVa | 0 to 100 |
| Professional | FULVa | 0 to 100 |
| Ethical | FULVa | 0 to 100 |