Moral Reasoning on Moral ID
85.51AccuracyGRPO+AIR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRPO+AIR2026.05 | 85.51 | 59.85 | |
| GSPO+AIR2026.05 | 84.57 | 65.69 | |
| GPT-5.2-chat-latest2026.05 | 83.21 | 29.2 | |
| GSPO2026.05 | 81.02 | 40.15 | |
| GPT-5-chat-latest2026.05 | 80.81 | 22.63 | |
| GRPO2026.05 | 75.39 | 34.31 | |
| Gemini-3-Pro-Preview2026.05 | 67.78 | 10.22 | |
| Qwen-3-14B2026.05 | 67.36 | 10.95 | |
| Llama-4-Maverick2026.05 | 64.96 | 5.11 | |
| Qwen-2.5-14B-Instruct2026.05 | 62.15 | 3.65 | |
| GSPO+V-REx2026.05 | 61.5 | 10.8 | |
| GRPO+V-REx2026.05 | 58.2 | 7.15 | |
| Gemini-2.5-Pro2026.05 | 57.51 | 7.3 |