List your company
ResearchFormerly WizWand

Chinese General Knowledge Question Answering on C-Eval

91.82AccuracyDeepSeek-R1
28.69245.08161.4777.859Feb 9, 2026Mar 3, 2026Mar 26, 2026Apr 18, 2026May 10, 2026Jun 2, 2026Jun 25, 2026
Updated 4mo ago

Evaluation results

MethodLinks
DeepSeek-R1Model Size=671B, Mode=Thinking2026.0291.82
Qwen3Model Size=32B, Mode=Thinking2026.0287.33
Qwen3Model Size=14B, Mode=Thinking2026.0286.26
Qwen3Model Size=8B, Mode=Thinking2026.0283.45
Qwen3Model Size=32B, Mode=No-Thinking2026.0283.32
Teacher-30BInference Mode=thinking2026.0683.28
MirrorModel Size=8B, Mode=Thinking2026.0282.17
Teacher-8BInference Mode=thinking2026.0681.73
Qwen3Model Size=14B, Mode=No-Thinking2026.0281.04
GPT-4.1Mode=No-Thinking2026.0277.97
Qwen3Model Size=8B, Mode=No-Thinking2026.0277.91
MirrorModel Size=8B, Mode=No-Thinking2026.0277.08
InternVL 3.0 2BNumber of Parameters=2B, Input Modality=Text-Only2026.0373.12
OPD-T30BInference Mode=thinking2026.0666.73
InternVL 3.5 2BNumber of Parameters=2B, Input Modality=Text-Only2026.0363.66
Xuanwu VL-2BNumber of Parameters=2B, Input Modality=Text-Only2026.0362.18
OPD-T8BInference Mode=thinking2026.0659.9
RLInference Mode=thinking2026.0649.79
Base (SFT ckpt)Inference Mode=thinking2026.0631.12