Social Reasoning on MotiveBench
94AccuracyO3
Evaluation Results
| Method | Links | |
|---|---|---|
| O32026.03 | 94 | |
| GPT-4o2026.03 | 93 | |
| GPT-5Mode=COT2026.03 | 92.67 | |
| O3Mode=COT2026.03 | 92 | |
| GPT-52026.03 | 90.89 | |
| GPT-4oMode=COT2026.03 | 89.78 | |
| Qwen3-32BThinking=Disabled2026.03 | 89.22 | |
| SocialR1-8BAblation=only Rout2026.03 | 88.89 | |
| LLaMa3.1-70BMode=COT2026.03 | 88.67 | |
| Distill-Llama-70B2026.03 | 88.44 | |
| Qwen3-32B2026.03 | 88 | |
| SocialR1-8BAblation=w/o Rstruct2026.03 | 87.89 | |
| SocialR1-8BAblation=w/o Rlen2026.03 | 87.67 | |
| SocialR1-8BAblation=Full2026.03 | 87.56 | |
| DeepSeek-R12026.03 | 86.55 | |
| SocialR1-8BAblation=w/o Rcont2026.03 | 85.67 | |
| SocialR1-4BAblation=only Rout2026.03 | 85.33 | |
| SocialR1-4BAblation=Full2026.03 | 85 | |
| LLaMa3.1-70B2026.03 | 84.45 | |
| Qwen3-4B2026.03 | 84.44 | |
| SocialR1-4BAblation=w/o Rlen2026.03 | 83.11 | |
| SocialR1-4BAblation=w/o Rstruct2026.03 | 83.11 | |
| SocialR1-4BAblation=w/o Rcont2026.03 | 83 | |
| Qwen3-8BThinking=Disabled2026.03 | 79.33 | |
| Qwen3-4BThinking=Disabled2026.03 | 77.34 | |
| Qwen3-8B2026.03 | 68.78 |