Long-context Question Answering on LoCoMo (Dimensional Breakdown)
87.4Single-hop AccuracyFull Context
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Full ContextPipeline category=Full Context2026.06 | 87.4 | 69.9 | 51.7 | 57.3 | 74.9 | |
| KaLM-Embedding-Gemma3-12BPipeline category=NaiveRAG2026.06 | 87.2 | 62.8 | 47.7 | 52.1 | 72.3 | |
| EvoEmbedding-4BPipeline category=NaiveRAG2026.06 | 86.6 | 71.6 | 49.8 | 56.3 | 74.3 | |
| Qwen3-Embedding-8BPipeline category=NaiveRAG2026.06 | 82.2 | 60.6 | 41.1 | 53.1 | 67.9 | |
| MemoryOSPipeline category=Agentic Memory Systems2026.06 | 72.3 | 62.8 | 33 | 51 | 61 | |
| A-MEMPipeline category=Agentic Memory Systems2026.06 | 67.9 | 57.5 | 27.7 | 43.8 | 56.1 | |
| Mem0Pipeline category=Agentic Memory Systems2026.06 | 67.7 | 54.3 | 57 | 46.9 | 61.7 | |
| LightMemPipeline category=Agentic Memory Systems2026.06 | 67 | 45.8 | 76.3 | 76.8 | 72.6 | |
| all-MiniLM-L6-v2Pipeline category=NaiveRAG2026.06 | 29.8 | 20.9 | 40.6 | 49 | 39.1 |