Social Reasoning on TactfulToM
69.69AccuracyDeepSeek-R1
Evaluation Results
| Method | Links | |
|---|---|---|
| DeepSeek-R12026.03 | 69.69 | |
| O32026.03 | 68.6 | |
| GPT-5Mode=COT2026.03 | 65.47 | |
| O3Mode=COT2026.03 | 65.44 | |
| Distill-Llama-70B2026.03 | 62.02 | |
| LLaMa3.1-70BMode=COT2026.03 | 55.83 | |
| Qwen3-32B2026.03 | 55.75 | |
| LLaMa3.1-70B2026.03 | 55.1 | |
| GPT-4oMode=COT2026.03 | 52.81 | |
| GPT-4o2026.03 | 52.67 | |
| Qwen3-32BThinking=Disabled2026.03 | 51.93 | |
| SocialR1-8BAblation=Full2026.03 | 50.79 | |
| SocialR1-4BAblation=w/o Rstruct2026.03 | 50.01 | |
| SocialR1-4BAblation=w/o Rlen2026.03 | 49.41 | |
| SocialR1-4BAblation=Full2026.03 | 49.36 | |
| SocialR1-8BAblation=w/o Rlen2026.03 | 48.72 | |
| SocialR1-8BAblation=only Rout2026.03 | 48.6 | |
| GPT-52026.03 | 46.7 | |
| SocialR1-8BAblation=w/o Rcont2026.03 | 46.67 | |
| Qwen3-8B2026.03 | 46.55 | |
| SocialR1-4BAblation=w/o Rcont2026.03 | 46.09 | |
| Qwen3-4B2026.03 | 45.84 | |
| SocialR1-8BAblation=w/o Rstruct2026.03 | 45.58 | |
| Qwen3-8BThinking=Disabled2026.03 | 42.92 | |
| SocialR1-4BAblation=only Rout2026.03 | 42.65 | |
| Qwen3-4BThinking=Disabled2026.03 | 40.31 |