Person understanding and persistent memory on LoCoMo-Plus
25.5ScorePERSONATREE
Evaluation Results
| Method | Links | |
|---|---|---|
| PERSONATREEBackbone=GPT-5.4 Mini2026.06 | 25.5 | |
| Flat retrievalBackbone=Gemini 3 Flash2026.06 | 25.2 | |
| Full historyBackbone=Gemini 3 Flash2026.06 | 24.8 | |
| PERSONATREEBackbone=Gemini 3 Flash2026.06 | 24.6 | |
| TiMemBackbone=GPT-5.4 Mini2026.06 | 24.2 | |
| TiMemBackbone=Gemini 3 Flash2026.06 | 23.6 | |
| Flat retrievalBackbone=GPT-5.4 Mini2026.06 | 23.6 | |
| Full historyBackbone=GPT-5.4 Mini2026.06 | 22.9 | |
| PERSONATREEBackbone=Qwen3-32B2026.06 | 22.3 | |
| TiMemBackbone=Qwen3-32B2026.06 | 20.9 | |
| A-MEMBackbone=GPT-5.4 Mini2026.06 | 20.7 | |
| Flat retrievalBackbone=Qwen3-32B2026.06 | 20.4 | |
| A-MEMBackbone=Gemini 3 Flash2026.06 | 20.3 | |
| Full historyBackbone=Qwen3-32B2026.06 | 20.1 | |
| Mem0Backbone=GPT-5.4 Mini2026.06 | 18.8 | |
| Mem0Backbone=Gemini 3 Flash2026.06 | 18.3 | |
| Benchmark strongBackbone=GPT-5.4 Mini2026.06 | 18.2 | |
| Benchmark strongBackbone=Gemini 3 Flash2026.06 | 17.8 | |
| A-MEMBackbone=Qwen3-32B2026.06 | 17.7 | |
| Mem0Backbone=Qwen3-32B2026.06 | 16.1 | |
| Benchmark strongBackbone=Qwen3-32B2026.06 | 15.3 |