Social Intelligence Assessment on SOTOPIA hard episodes (test)
7.21GOA ScoreGPT-5
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GPT-52026.03 | 7.21 | — | — | — | — | — | — | — | |
| RAPOBackbone=Qwen2.5-7B-Instruct2026.03 | 7.19 | — | — | — | — | — | — | 3.54 | |
| CPOBackbone=Qwen2.5-7B-Instruct2026.03 | 7.14 | — | — | — | — | — | — | 3.48 | |
| GRPO iStarBackbone=Qwen2.5-7B-Instruct2026.03 | 7.11 | — | — | — | — | — | — | — | |
| GPT-4o2026.03 | 6.97 | — | — | — | — | — | — | 3.46 | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.03 | 6.97 | — | — | — | — | — | — | — | |
| PPOBackbone=Qwen2.5-7B-Instruct2026.03 | 6.63 | — | — | — | — | — | — | — | |
| Doubao-Pro-32k2026.03 | 6.49 | — | — | — | — | — | — | 3.22 | |
| DEMO-Qwen2-7BBackbone=Qwen2-7B, Method=DEMO Agent (Expert Experience Imitation)2024.12 | 6.4 | -0.02 | 0 | 0.82 | 2.32 | 4.52 | 8.94 | 3.28 | |
| Claude-3.5-Sonnet2026.03 | 6.33 | — | — | — | — | — | — | 3.09 | |
| Qwen2.5-7B-Instruct2026.03 | 6.21 | — | — | — | — | — | — | 3.01 | |
| Qwen2-7B-InstructBackbone=Qwen2-7B, Variant=Instruct2024.12 | 6.07 | -0.05 | 0 | 0.73 | 1.83 | 3.41 | 8.64 | 2.95 | |
| DeepSeek-R12026.03 | 5.86 | — | — | — | — | — | — | 2.73 | |
| RLOOBackbone=Qwen2.5-7B-Instruct2026.03 | 5.7 | — | — | — | — | — | — | — | |
| DEMO-Llama3.1-8BBackbone=Llama3.1-8B, Method=DEMO Agent (Expert Experience Imitation)2024.12 | 3.77 | -0.19 | 0 | 0.29 | 0.85 | 2.88 | 8.41 | 2.29 | |
| Llama3.1-8B-InstructBackbone=Llama3.1-8B, Variant=Instruct2024.12 | 3.39 | -0.5 | -0.01 | -0.16 | 0.6 | 2.21 | 8.63 | 1.85 |