Long-context evaluation on LB v2 (ALL)
38Accuracy (ALL)RAG
Evaluation Results
| Method | Links | |
|---|---|---|
| RAGContext Length=32k2026.01 | 38 | |
| RR+Judge(y)Context Type=Full Context, Max Length=1M tokens2026.01 | 32.6 | |
| BaseContext Type=Full Context, Max Length=1M tokens2026.01 | 32 | |
| RID+Q(y)Context Type=Full Context, Max Length=1M tokens2026.01 | 32 | |
| RID+Q(y)Context Type=KV-Cache with RetrievalAttn2026.01 | 32 | |
| RID+C(y)Context Type=Full Context, Max Length=1M tokens2026.01 | 31.4 | |
| RAO(y)Context Type=Full Context, Max Length=1M tokens2026.01 | 31.2 | |
| RR+Judge(y)Context Type=KV-Cache with RetrievalAttn2026.01 | 31.2 | |
| RID(y)Context Type=Full Context, Max Length=1M tokens2026.01 | 31 | |
| BaseContext Type=KV-Cache with RetrievalAttn2026.01 | 30.2 | |
| RID(y)Context Type=KV-Cache with RetrievalAttn2026.01 | 30 | |
| RAO(y)Context Type=KV-Cache with RetrievalAttn2026.01 | 29.6 | |
| RID+C(y)Context Type=KV-Cache with RetrievalAttn2026.01 | 29.6 |