List your company
ResearchFormerly WizWand

Social Interaction on SOTOPIA all social scenarios

8.95Goal ScoreGPT-5
4.91485.96247.018.0576Mar 13, 2024Jul 19, 2024Nov 24, 2024Apr 1, 2025Aug 7, 2025Dec 13, 2025Apr 21, 2026
Updated 6mo ago

Evaluation results

MethodLinks
GPT-52026.038.95——
SAVOIREvaluation Setting=Self-Play, Base Model=Qwen2.5-7B-Instruct2026.048.43—3.85
GRPO iStarBackbone=Qwen2.5-7B-Instruct2026.038.42——
Claude-3.5-SonnetEvaluation Setting=GPT-4o-as-Partner2026.048.42—3.77
SAVOIREvaluation Setting=GPT-4o-as-Partner, Base Model=Qwen2.5-7B-Instruct2026.048.42—3.94
RAPOBackbone=Qwen2.5-7B-Instruct2026.038.413.87—
EPOEvaluation Setting=GPT-4o-as-Partner, Base Model=Qwen2.5-7B-Instruct2026.048.41—3.86
DSIEvaluation Setting=Self-Play, Base Model=Qwen2.5-7B-Instruct2026.048.35—3.75
GRPOBackbone=Qwen2.5-7B-Instruct2026.038.31——
Sotopia-RLEvaluation Setting=GPT-4o-as-Partner, Base Model=Qwen2.5-7B-Instruct2026.048.31—3.9
Claude-3.5-Sonnet2026.038.293.71—
Claude-3.5-SonnetEvaluation Setting=Self-Play2026.048.29—3.71
PPOBackbone=Qwen2.5-7B-Instruct2026.038.25——
GPT-4o2026.038.193.76—
GPT-4oEvaluation Setting=Self-Play2026.048.19—3.76
GPT-4oEvaluation Setting=GPT-4o-as-Partner2026.048.19—3.76
DeepSeek-V3Evaluation Setting=Self-Play2026.048.15—3.62
DSIEvaluation Setting=GPT-4o-as-Partner, Base Model=Qwen2.5-7B-Instruct2026.048.15—3.7
DeepSeek-V3Evaluation Setting=GPT-4o-as-Partner2026.048.14—3.72
CPOBackbone=Qwen2.5-7B-Instruct2026.038.133.77—
Doubao-Pro-32k2026.038.123.7—
Gemini-2.5-ProEvaluation Setting=GPT-4o-as-Partner2026.048.12—3.59
DATEvaluation Setting=GPT-4o-as-Partner, Base Model=Qwen2.5-7B-Instruct2026.048.11—3.7
EPOEvaluation Setting=Self-Play, Base Model=Qwen2.5-7B-Instruct2026.048.09—3.51
OpenAI-o1Evaluation Setting=GPT-4o-as-Partner2026.048.09—3.69
PPDPPEvaluation Setting=GPT-4o-as-Partner, Base Model=Qwen2.5-7B-Instruct2026.048.07—3.71
DeepSeek-R12026.037.973.4—
DeepSeek-R1Evaluation Setting=Self-Play2026.047.97—3.4
PPDPPEvaluation Setting=Self-Play, Base Model=Qwen2.5-7B-Instruct2026.047.97—3.65
DATEvaluation Setting=Self-Play, Base Model=Qwen2.5-7B-Instruct2026.047.97—3.59
OpenAI-o3-miniEvaluation Setting=GPT-4o-as-Partner2026.047.96—3.61
OpenAI-o1Evaluation Setting=Self-Play2026.047.93—3.58
DeepSeek-R1Evaluation Setting=GPT-4o-as-Partner2026.047.92—3.49
Qwen2.5-7B-Instruct2026.037.913.55—
Qwen2.5-7B-InstructEvaluation Setting=Self-Play2026.047.91—3.55
Gemini-2.5-ProEvaluation Setting=Self-Play2026.047.85—3.43
Sotopia-RLEvaluation Setting=Self-Play, Base Model=Qwen2.5-7B-Instruct2026.047.8—3.55
QwQ-32BEvaluation Setting=GPT-4o-as-Partner2026.047.8—3.47
QwQ-32BEvaluation Setting=Self-Play2026.047.7—3.3
Expert (GPT-4)Backbone=GPT-42024.037.623.31—
BC+SRBackbone=Mistral-7B, Method=SOTOPIA-π2024.037.623.44—
OpenAI-o3-miniEvaluation Setting=Self-Play2026.047.38—3.3
Behavior Cloning (BC)Backbone=Mistral-7B, Method=SOTOPIA-π2024.037.273.41—
RLOOBackbone=Qwen2.5-7B-Instruct2026.037.13——
Qwen2.5-7B-InstructEvaluation Setting=GPT-4o-as-Partner2026.046.71—3.13
Self-Reinforcement (SR)Backbone=Mistral-7B, Method=SOTOPIA-π2024.035.832.57—
Base (Mistral-7B)Backbone=Mistral-7B2024.035.072.33—