Retrieval-Augmented Question Answering on 170 unique canonical user-facing queries Simple Rec. n=50
96FaithfulnessGPT-5.5
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5.52026.06 | 96 | — | 60 | 57 | |
| GPT-4o2026.06 | 95 | — | 48 | 50 | |
| Qwen2.52026.06 | 94 | — | 37 | 44 | |
| I-GUIDE SSmode=Smart Search2026.06 | 75 | 63 | 76 | 70 | |
| Naive RAG2026.06 | 67 | 17 | 30 | 23 | |
| Claude Opus 4.72026.06 | 66 | — | 43 | 35 |