Streaming Long-context Question Answering on STEM-Bench The Office
0.304TimeProStream (-w/o Mpend)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ProStream (-w/o Mpend)ablation=Eliminating Pending buffer2026.03 | 0.304 | 11.1 | 40.1 | 81.7 | 48.7 | 61.7 | 52.8 | |
| ProStream2026.03 | 0.316 | 11.4 | 40.8 | 81.9 | 49.1 | 62.6 | 53.4 | |
| ProStream (-w/o Mtree)ablation=Removing the Hierarchical Tree2026.03 | 0.319 | 11 | 40.2 | 81.7 | 48.4 | 62 | 52.6 | |
| RAPTOR2026.03 | 0.443 | 3.2 | 14.2 | 74.5 | 32.4 | 33.4 | 20.1 | |
| ProStream (-w/o B)ablation=Eliminating the Short-Term Sensing Buffer2026.03 | 0.452 | 3 | 13.8 | 75.8 | 19.1 | 33.6 | 19.3 | |
| RAG2026.03 | 0.918 | 7.4 | 27.1 | 77.8 | 32.2 | 52.4 | 38.5 | |
| RQ-RAG2026.03 | 1.058 | 7.2 | 27 | 78 | 46.6 | 53.5 | 37.6 | |
| GraphRAG2026.03 | 1.231 | 5.5 | 22.3 | 77.9 | 42 | 43.3 | 31.1 | |
| A-Mem2026.03 | 1.272 | 5.9 | 22 | 75.3 | 35.3 | 37.7 | 29.5 | |
| Summary2026.03 | 1.641 | 4.8 | 20.1 | 77 | 36.5 | 39.7 | 27.2 | |
| HippoRAG22026.03 | 4.404 | 4 | 16.9 | 74 | 49.9 | 45.8 | 27 | |
| Full Context2026.03 | 4.776 | 6.6 | 26.6 | 74.2 | 29.2 | 73.1 | 57.4 | |
| MemoRAG2026.03 | 5.922 | 3.6 | 17.1 | 77.3 | 56.8 | 34.5 | 22.5 | |
| MemGAS2026.03 | 7.694 | 0.6 | 3.9 | 59.2 | 54.7 | 41.7 | 28.5 |