List your company

Chinese Idiom Cloze Completion on CHID (held-out)

70.3DBSSOURCEbest
1.24419.17237.155.028Feb 11, 2026

Evaluation results

MethodLinks
SOURCEbest2026.0270.3—
EXPERTModel=Qwen3-1.7B optimized2026.0259.8—
SOURCEavg2026.0249.3—
DataChef-32BOracle Upper Bound=true, Samples=322026.0245.7—
Gemini-3-Pro2026.0221.929.7
DataChef-32B2026.0220.57.9
Qwen3-32B2026.0212.31.1
Qwen3-Next ⊕ Kimi-K2Reasoning backbone=Qwen3-Next-80B, Coding backbone=Kimi-K2-Instruct2026.024.18.6
Kimi-K22026.023.923.1