Long-context Classification on OOLONG (test)
58.1TREC-Q-coarse AccuracyRLM + PEEK
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RLM + PEEKBase LM=GPT-5-mini-2025-08-072026.05 | 58.1 | 69.4 | 57 | |
| RLM + ACE (Online Adaptation)Base LM=GPT-5-mini-2025-08-072026.05 | 48.8 | 61.6 | 42 | |
| RLM + Compaction AgentBase LM=GPT-5-mini-2025-08-072026.05 | 42 | 49.5 | 30 | |
| RLM + RAGBase LM=GPT-5-mini-2025-08-072026.05 | 36.6 | 63.1 | 29 | |
| RLM + Shared ChatBase LM=GPT-5-mini-2025-08-072026.05 | 32 | 49.6 | 23 | |
| RLMBase LM=GPT-5-mini-2025-08-072026.05 | 30.3 | 46.5 | 23 |