Key-value recall on Memory KV 37 cases (test)
37AccuracyGPT-5
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5Gen. Traj.=N/A2026.06 | 37 | |
| Qwen3-4B-ThinkingGen. Traj.=N/A2026.06 | 34 | |
| Qwen3-4B-Think + LoRA (1000)Gen. Traj.=✓2026.06 | 34 | |
| Qwen3-4B-Think + LoRA (3689)Gen. Traj.=×2026.06 | 31 | |
| Qwen3-4B-Think + Full (1000)Gen. Traj.=✓2026.06 | 29 | |
| Qwen3-4B-Think + Full (3689)Gen. Traj.=×2026.06 | 28 | |
| Qwen3-4B-InstructGen. Traj.=N/A2026.06 | 17 |