Kindness
How warmly each model responds to human-centered prompts.
| Model | Win Rate | BT Elo |
|---|---|---|
1anthropic/claude-3-5-haiku | 100% | 1620 |
2GPT-4o Mini | 100% | 1620 |
3Llama 3.3 70B | 0% | 300 |
4Mistral Small 3.1 | 0% | 300 |
Leaderboard
How frontier models rank on human values.
Rankings are built from human pairwise votes. In each battle, two models respond to the same prompt — voters choose which response better embodies the target value. Bradley-Terry scoring converts those votes into Elo ratings across four dimensions.
EigenBench rankings
Bradley-Terry Elo computed from human pairwise votes.
Updated every 5 min · live from votes
Conservatism
Preference for stability, institutions, and incremental change.
| Model | Win Rate | BT Elo |
|---|---|---|
1Llama 3.3 70B | 200% | 1741 |
2Gemini 2.0 Flash | 50% | 300 |
3GPT-4o Mini | 50% | 300 |
4Claude 3.5 Haiku | 100% | 300 |
Deep Ecology
Alignment with ecological stewardship and planet-first values.
Not enough votes yet — keep judging to unlock rankings.
Loyalty
Commitment to sustained relationships and group solidarity.
| Model | Win Rate | BT Elo |
|---|---|---|
1Qwen 2.5 72B | 100% | 1570 |
2Claude 3.5 Haiku | 100% | 1570 |
3Mistral Small 3.1 | 0% | 300 |
Cross-dimension performance
How each model ranks across all four value dimensions. Darker = higher ranked.
| Model | Kindness | Conservatism | Deep Ecology | Loyalty | Avg Rank |
|---|---|---|---|---|---|
| Qwen 2.5 72B | — | — | — | #1 1570 | 1.0 |
| Gemini 2.0 Flash | — | #2 300 | — | — | 2.0 |
| Llama 3.3 70B | #3 300 | #1 1741 | — | — | 2.0 |
| GPT-4o Mini | #2 1620 | #3 300 | — | — | 2.5 |
| Claude 3.5 Haiku | — | #4 300 | — | #2 1570 | 3.0 |
| Mistral Small 3.1 | #4 300 | — | — | #3 300 | 3.5 |
Head-to-head results
Win rate of the row model vs each column opponent. Ties are split 50/50. Darker = higher win rate.
Kindness
Row wins vs column
| 4oMi | Haik | Gmni | Llma | Mist | Qwen | |
|---|---|---|---|---|---|---|
| GPT-4o Mini | — | — | — | 100% 1v | 100% 1v | — |
| Claude 3.5 Haiku | — | — | — | — | — | — |
| Gemini 2.0 Flash | — | — | — | — | — | — |
| Llama 3.3 70B | 0% 1v | — | — | — | — | — |
| Mistral Small 3.1 | 0% 1v | — | — | — | — | — |
| Qwen 2.5 72B | — | — | — | — | — | — |
Conservatism
Row wins vs column
| 4oMi | Haik | Gmni | Llma | Mist | Qwen | |
|---|---|---|---|---|---|---|
| GPT-4o Mini | — | 50% 2v | 0% 1v | — | — | — |
| Claude 3.5 Haiku | 50% 2v | — | — | — | — | — |
| Gemini 2.0 Flash | 100% 1v | — | — | 0% 2v | — | — |
| Llama 3.3 70B | — | — | 100% 2v | — | — | — |
| Mistral Small 3.1 | — | — | — | — | — | — |
| Qwen 2.5 72B | — | — | — | — | — | — |
Deep Ecology
Row wins vs column
| 4oMi | Haik | Gmni | Llma | Mist | Qwen | |
|---|---|---|---|---|---|---|
| GPT-4o Mini | — | — | — | — | — | — |
| Claude 3.5 Haiku | — | — | — | — | — | — |
| Gemini 2.0 Flash | — | — | — | — | — | — |
| Llama 3.3 70B | — | — | — | — | — | — |
| Mistral Small 3.1 | — | — | — | — | — | — |
| Qwen 2.5 72B | — | — | — | — | — | — |
Loyalty
Row wins vs column
| 4oMi | Haik | Gmni | Llma | Mist | Qwen | |
|---|---|---|---|---|---|---|
| GPT-4o Mini | — | — | — | — | — | — |
| Claude 3.5 Haiku | — | — | — | — | 100% 1v | — |
| Gemini 2.0 Flash | — | — | — | — | — | — |
| Llama 3.3 70B | — | — | — | — | — | — |
| Mistral Small 3.1 | — | 0% 1v | — | — | — | 0% 1v |
| Qwen 2.5 72B | — | — | — | — | 100% 1v | — |
Contribute
Your votes shape the rankings.
Head to the battle page, read two model responses side by side, and pick which one better reflects the target value. Every vote updates the leaderboard.