Streaming Long-context Question Answering on STEM-Bench The Big Bang Theory
0.4Time (s)ProStream (-w/o Mtree)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ProStream (-w/o Mtree)ablation=Removing the Hierarchical Tree2026.03 | 0.4 | 0.116 | 0.382 | 0.81 | 46.3 | 0.607 | 0.5 | |
| ProStream (-w/o Mpend)ablation=Eliminating Pending buffer2026.03 | 0.401 | 0.115 | 0.38 | 0.81 | 47.5 | 0.602 | 0.489 | |
| ProStream2026.03 | 0.408 | 0.118 | 0.39 | 0.817 | 48.5 | 0.612 | 0.509 | |
| ProStream (-w/o B)ablation=Eliminating the Short-Term Sensing Buffer2026.03 | 0.409 | 0.049 | 0.182 | 0.783 | 7.1 | 0.369 | 0.255 | |
| RAPTOR2026.03 | 0.519 | 0.041 | 0.165 | 0.761 | 30.2 | 0.375 | 0.214 | |
| RAG2026.03 | 0.957 | 0.068 | 0.247 | 0.781 | 35.5 | 0.508 | 0.331 | |
| GraphRAG2026.03 | 0.997 | 0.069 | 0.246 | 0.788 | 41.7 | 0.479 | 0.341 | |
| A-Mem2026.03 | 1.301 | 0.057 | 0.204 | 0.755 | 37 | 0.371 | 0.272 | |
| Summary2026.03 | 2.001 | 0.049 | 0.2 | 0.783 | 29.6 | 0.407 | 0.278 | |
| HippoRAG22026.03 | 4.231 | 0.046 | 0.176 | 0.742 | 48.1 | 0.48 | 0.263 | |
| RQ-RAG2026.03 | 5.149 | 0.051 | 0.201 | 0.77 | 27.5 | 0.46 | 0.329 | |
| MemGAS2026.03 | 5.711 | 0.013 | 0.07 | 0.528 | 40.2 | 0.768 | 0.508 | |
| Full Context2026.03 | 7.333 | 0.065 | 0.249 | 0.737 | 25 | 0.71 | 0.521 | |
| MemoRAG2026.03 | 7.498 | 0.049 | 0.193 | 0.791 | 55.1 | 0.366 | 0.265 |