Person understanding and persistent memory on RealMem
49.8ScorePERSONATREE
Evaluation Results
| Method | Links | |
|---|---|---|
| PERSONATREEBackbone=GPT-5.4 Mini2026.06 | 49.8 | |
| Benchmark strongBackbone=GPT-5.4 Mini2026.06 | 48.2 | |
| PERSONATREEBackbone=Qwen3-32B2026.06 | 47.8 | |
| Benchmark strongBackbone=Gemini 3 Flash2026.06 | 47.4 | |
| PERSONATREEBackbone=Gemini 3 Flash2026.06 | 47.1 | |
| Mem0Backbone=GPT-5.4 Mini2026.06 | 46.6 | |
| Benchmark strongBackbone=Qwen3-32B2026.06 | 46.2 | |
| TiMemBackbone=GPT-5.4 Mini2026.06 | 46.1 | |
| Mem0Backbone=Gemini 3 Flash2026.06 | 45.6 | |
| TiMemBackbone=Gemini 3 Flash2026.06 | 45.1 | |
| Mem0Backbone=Qwen3-32B2026.06 | 44.4 | |
| TiMemBackbone=Qwen3-32B2026.06 | 44.3 | |
| A-MEMBackbone=GPT-5.4 Mini2026.06 | 42.9 | |
| Flat retrievalBackbone=GPT-5.4 Mini2026.06 | 42.4 | |
| A-MEMBackbone=Gemini 3 Flash2026.06 | 41.6 | |
| Flat retrievalBackbone=Gemini 3 Flash2026.06 | 41.5 | |
| Full historyBackbone=GPT-5.4 Mini2026.06 | 41.3 | |
| A-MEMBackbone=Qwen3-32B2026.06 | 40.5 | |
| Full historyBackbone=Gemini 3 Flash2026.06 | 40.4 | |
| Flat retrievalBackbone=Qwen3-32B2026.06 | 40.3 | |
| Full historyBackbone=Qwen3-32B2026.06 | 39.1 |