Moral Reasoning on Moral OOD
81.07AccuracyGSPO+AIR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GSPO+AIR2026.05 | 81.07 | 49.26 | |
| GRPO+AIR2026.05 | 80.7 | 47.79 | |
| GPT-5.2-chat-latest2026.05 | 77.39 | 25 | |
| GPT-5-chat-latest2026.05 | 77.21 | 30.15 | |
| Qwen-3-14B2026.05 | 72.61 | 25.74 | |
| Gemini-3-Pro-Preview2026.05 | 68.2 | 12.5 | |
| Llama-4-Maverick2026.05 | 68.01 | 10.29 | |
| GRPO2026.05 | 62.68 | 14.71 | |
| GSPO2026.05 | 61.03 | 19.12 | |
| Gemini-2.5-Pro2026.05 | 60.48 | 11.03 | |
| GSPO+V-REx2026.05 | 59.74 | 11.76 | |
| Qwen-2.5-14B-Instruct2026.05 | 58.09 | 9.56 | |
| GRPO+V-REx2026.05 | 53.12 | 3.68 |