Social Reasoning on Hi-ToM
70.83AccuracySocialR1-8B
Evaluation Results
| Method | Links | |
|---|---|---|
| SocialR1-8BAblation=Full2026.03 | 70.83 | |
| SocialR1-8BAblation=w/o Rstruct2026.03 | 70.67 | |
| DeepSeek-R12026.03 | 70.59 | |
| O32026.03 | 70.33 | |
| SocialR1-8BAblation=w/o Rcont2026.03 | 69.5 | |
| GPT-5Mode=COT2026.03 | 67.5 | |
| O3Mode=COT2026.03 | 66.5 | |
| SocialR1-4BAblation=Full2026.03 | 65.83 | |
| SocialR1-4BAblation=w/o Rlen2026.03 | 64 | |
| GPT-52026.03 | 63.17 | |
| SocialR1-8BAblation=w/o Rlen2026.03 | 62.67 | |
| SocialR1-8BAblation=only Rout2026.03 | 62.33 | |
| GPT-4oMode=COT2026.03 | 60.5 | |
| GPT-4o2026.03 | 60.33 | |
| LLaMa3.1-70B2026.03 | 59.83 | |
| LLaMa3.1-70BMode=COT2026.03 | 58.5 | |
| SocialR1-4BAblation=w/o Rcont2026.03 | 58.5 | |
| Qwen3-4BThinking=Disabled2026.03 | 57 | |
| Qwen3-4B2026.03 | 56.17 | |
| SocialR1-4BAblation=w/o Rstruct2026.03 | 56.17 | |
| Qwen3-8BThinking=Disabled2026.03 | 56 | |
| Qwen3-32B2026.03 | 55.17 | |
| Qwen3-8B2026.03 | 55 | |
| Qwen3-32BThinking=Disabled2026.03 | 55 | |
| Distill-Llama-70B2026.03 | 53.33 | |
| SocialR1-4BAblation=only Rout2026.03 | 52.83 |