Long-context Question Answering on NQ (Exact Match)
56Exact Match (EM)Codex (No Retriever)
Evaluation Results
| Method | Links | |
|---|---|---|
| Codex (No Retriever)Retriever=None, Base Model=GPT-52026.03 | 56 | |
| RLM2026.03 | 55.33 | |
| Codex + BM25Retriever=BM25, Base Model=GPT-52026.03 | 53 | |
| Best Published2026.03 | 50.9 | |
| ReAct AgentAgent Loop=ReAct2026.03 | 49 | |
| RAGRetriever=BM252026.03 | 47 | |
| GPT-5 Full ContextContext Type=Full Context2026.03 | 27 |