Chinese Idiom Cloze Completion on CHID (held-out)
70.3DBSSOURCEbest
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SOURCEbest2026.02 | 70.3 | — | |
| EXPERTModel=Qwen3-1.7B optimized2026.02 | 59.8 | — | |
| SOURCEavg2026.02 | 49.3 | — | |
| DataChef-32BOracle Upper Bound=true, Samples=322026.02 | 45.7 | — | |
| Gemini-3-Pro2026.02 | 21.9 | 29.7 | |
| DataChef-32B2026.02 | 20.5 | 7.9 | |
| Qwen3-32B2026.02 | 12.3 | 1.1 | |
| Qwen3-Next ⊕ Kimi-K2Reasoning backbone=Qwen3-Next-80B, Coding backbone=Kimi-K2-Instruct2026.02 | 4.1 | 8.6 | |
| Kimi-K22026.02 | 3.9 | 23.1 |