Cultural Question Answering on BLEnD MCQ subset
68.18Accuracy (low-su)CORAL (GPT-OSS-120B)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CORAL (GPT-OSS-120B)Generator Model=LLAMA-3.2-3B-INSTRUCT, RAG Strategy=CORAL, Planner/Critic Model=GPT-OSS-120B2026.04 | 68.18 | 60.47 | 70.92 | 69.1 | 74.36 | 73.51 | 67.14 | |
| CORAL (QWEN3-235B)Generator Model=LLAMA-3.2-3B-INSTRUCT, RAG Strategy=CORAL, Planner/Critic Model=QWEN3-235B2026.04 | 66.78 | 61.83 | 72.22 | 70.41 | 71.93 | 72.76 | 67.84 | |
| crossRAGGenerator Model=LLAMA-3.2-3B-INSTRUCT, RAG Strategy=crossRAG2026.04 | 62.59 | 57.83 | 67.32 | 66.83 | 68.29 | 69.76 | 64.27 | |
| multiRAGGenerator Model=LLAMA-3.2-3B-INSTRUCT, RAG Strategy=multiRAG2026.04 | 61.89 | 56.48 | 67.97 | 65.92 | 67.98 | 69.84 | 63.49 | |
| Non-RAGGenerator Model=LLAMA-3.2-3B-INSTRUCT, RAG Strategy=None2026.04 | 58.04 | 55.65 | 62.09 | 63.06 | 68.59 | 69.29 | 62.13 | |
| monoRAGGenerator Model=LLAMA-3.2-3B-INSTRUCT, RAG Strategy=monoRAG2026.04 | 57.69 | 56.8 | 65.03 | 65.47 | 68.44 | 71.31 | 63.93 |