Memory-augmented question answering on EngramaBench full v1
73.39Accuracy (Single)GPT-4o full-context
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| GPT-4o full-contextAnswering model=GPT-4o, Memory architecture=brute-force context inclusion2026.04 | 73.39 | 62.91 | 39.02 | 97.5 | 33.05 | 61.86 | 3.33 | |
| Engrama fullAnswering model=GPT-4o, Memory architecture=structured memory2026.04 | 59.97 | 65.32 | 32.3 | 80 | 29.63 | 53.67 | 0.67 | |
| Mem0Answering model=GPT-4o2026.04 | 28.48 | 52.66 | 23.56 | 100 | 22.55 | 48.09 | 0.36 |