Model Editing on UltraEditBench
88.88EfficacySTABLEEDIT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| STABLEEDITBackbone=Llama-3-8B-Instruct2026.05 | 88.88 | 85.46 | 69.06 | |
| UltraEditBackbone=LLaMA-3-8B-Instruct, Number of Edits=20K2025.05 | 85.7 | 81.28 | 68.73 | |
| UltraEditBackbone=LLaMA-3-8B-Instruct, Editing Instances=5002025.05 | 85.7 | 81.28 | 68.73 | |
| ULTRAEDITBackbone=Llama-3-8B-Instruct2026.05 | 85.7 | 81.28 | 68.73 | |
| RLEditBackbone=LLaMA-3-8B-Instruct, Number of Edits=20K2025.05 | 85.69 | 81.88 | 65.64 | |
| STABLEEDITBackbone=Mistral-7B-v0.32026.05 | 84.98 | 78.77 | 68.36 | |
| UltraEditBackbone=GPT-J, Number of Edits=20K2025.05 | 84.03 | 76.62 | 64.03 | |
| UltraEditBackbone=GPT-J, Editing Instances=5002025.05 | 84.03 | 76.62 | 64.03 | |
| ULTRAEDITBackbone=GPT-J-6B2026.05 | 84.03 | 76.62 | 64.03 | |
| UltraEditBackbone=Mistral-7B-v0.3, Number of Edits=20K2025.05 | 83.71 | 77.3 | 67.26 | |
| UltraEditBackbone=Mistral-7B-v0.3, Editing Instances=5002025.05 | 83.71 | 77.3 | 67.26 | |
| ULTRAEDITBackbone=Mistral-7B-v0.32026.05 | 83.71 | 77.3 | 67.26 | |
| UltraEdit*Backbone=LLaMA-3-8B-Instruct, Number of Edits=2M2025.05 | 83.45 | 79.11 | 78.05 | |
| UltraEditBackbone=LLaMA-3-8B-Instruct, Editing Instances=10002025.05 | 83.45 | 79.11 | 78.05 | |
| STABLEEDITBackbone=GPT-J-6B2026.05 | 83.41 | 76.55 | 67.17 | |
| RLEDITBackbone=Llama-3-8B-Instruct2026.05 | 83.37 | 78.65 | 63.31 | |
| UltraEdit*Backbone=Mistral-7B-v0.3, Number of Edits=2M2025.05 | 81.7 | 77.25 | 77.09 | |
| UltraEditBackbone=Mistral-7B-v0.3, Editing Instances=10002025.05 | 81.7 | 77.25 | 77.09 | |
| UltraEdit*Backbone=GPT-J, Number of Edits=2M2025.05 | 81.65 | 76.8 | 76.44 | |
| UltraEditBackbone=GPT-J, Editing Instances=10002025.05 | 81.65 | 76.8 | 76.44 | |
| RLEditBackbone=GPT-J, Number of Edits=20K2025.05 | 81.42 | 75.35 | 62.13 | |
| RLEDITBackbone=GPT-J-6B2026.05 | 81.14 | 74.88 | 61.87 | |
| UltraEdit*Backbone=Qwen2.5-7B-Instruct, Number of Edits=2M2025.05 | 80.7 | 75.78 | 76.01 | |
| UltraEditBackbone=Qwen2.5-7B-Instruct, Editing Instances=10002025.05 | 80.7 | 75.78 | 76.01 | |
| UltraEditBackbone=Qwen2.5-7B-Instruct, Number of Edits=20K2025.05 | 79.01 | 71.45 | 64.1 | |
| UltraEditBackbone=Qwen2.5-7B-Instruct, Editing Instances=5002025.05 | 79.01 | 71.45 | 64.1 | |
| RLEditBackbone=Mistral-7B-v0.3, Number of Edits=20K2025.05 | 76.5 | 70.68 | 61.29 | |
| RLEDITBackbone=Mistral-7B-v0.32026.05 | 69.85 | 65.1 | 57.08 | |
| WISEBackbone=GPT-J, Number of Edits=20K2025.05 | 52.51 | 47.88 | 47.5 | |
| WISEBackbone=Mistral-7B-v0.3, Number of Edits=20K2025.05 | 47.21 | 44.55 | 39.13 | |
| RLEditBackbone=Qwen2.5-7B-Instruct, Number of Edits=20K2025.05 | 47.08 | 38.76 | 39.27 | |
| WISEBackbone=LLaMA-3-8B-Instruct, Number of Edits=20K2025.05 | 42.27 | 41.65 | 39.79 | |
| MALMENBackbone=LLaMA-3-8B-Instruct2025.05 | 40.64 | 34.18 | 37.29 | |
| MALMENBackbone=Llama-3-8B-Instruct2026.05 | 40.64 | 34.18 | 37.29 | |
| Pre-editedBackbone=Mistral-7B-v0.32026.05 | 30.83 | 29.94 | 31.11 | |
| Pre-editedBackbone=Llama-3-8B-Instruct2026.05 | 27.29 | 26.4 | 27.24 | |
| AlphaEditBackbone=GPT-J, Number of Edits=20K2025.05 | 22.19 | 12.09 | 6.88 | |
| ALPHAEDITBackbone=GPT-J-6B2026.05 | 22.19 | 12.09 | 6.88 | |
| FTBackbone=GPT-J, Number of Edits=20K2025.05 | 22.03 | 17.61 | 19.6 | |
| FTBackbone=GPT-J-6B2026.05 | 22.03 | 17.61 | 19.6 | |
| Pre-editedBackbone=GPT-J-6B2026.05 | 22.01 | 21.47 | 22.2 | |
| AlphaEditBackbone=Qwen2.5-7B-Instruct, Number of Edits=20K2025.05 | 17.44 | 8.01 | 5.42 | |
| FTBackbone=LLaMA-3-8B-Instruct, Number of Edits=20K2025.05 | 13.5 | 11.92 | 10.18 | |
| FTBackbone=Llama-3-8B-Instruct2026.05 | 13.5 | 11.92 | 10.18 | |
| FTBackbone=Qwen2.5-7B-Instruct, Number of Edits=20K2025.05 | 13.2 | 10.53 | 10.27 | |
| AlphaEditBackbone=LLaMA-3-8B-Instruct, Number of Edits=20K2025.05 | 4.51 | 3.16 | 2.78 | |
| ALPHAEDITBackbone=Llama-3-8B-Instruct2026.05 | 4.51 | 3.16 | 2.78 | |
| MALMENBackbone=Qwen2.5-7B-Instruct2025.05 | 4.36 | 3.48 | 4.38 | |
| MENDBackbone=Qwen2.5-7B-Instruct2025.05 | 3.48 | 3.45 | 3.43 | |
| MALMENBackbone=Mistral-7B-v0.32025.05 | 2.42 | 2.48 | 3.47 | |
| MALMENBackbone=Mistral-7B-v0.32026.05 | 2.42 | 2.48 | 3.47 | |
| RECTBackbone=Qwen2.5-7B-Instruct2025.05 | 1.86 | 1.55 | 1.84 | |
| MENDBackbone=GPT-J2025.05 | 1.71 | 1.71 | 1.83 | |
| MENDBackbone=GPT-J-6B2026.05 | 1.71 | 1.71 | 1.83 | |
| PRUNEBackbone=LLaMA-3-8B-Instruct2025.05 | 1.21 | 0.85 | 0.32 | |
| MEMITBackbone=Qwen2.5-7B-Instruct2025.05 | 0.82 | 0.32 | 0.69 | |
| MALMENBackbone=GPT-J2025.05 | 0.76 | 0.48 | 0.78 | |
| MALMENBackbone=GPT-J-6B2026.05 | 0.76 | 0.48 | 0.78 | |
| MEMITBackbone=LLaMA-3-8B-Instruct2025.05 | 0.74 | 0.45 | 0.21 | |
| MEMITBackbone=Llama-3-8B-Instruct2026.05 | 0.74 | 0.45 | 0.21 | |
| RECTBackbone=LLaMA-3-8B-Instruct2025.05 | 0.55 | 0.09 | 0 | |
| PRUNEBackbone=GPT-J2025.05 | 0.32 | 0.27 | 0.28 | |
| PRUNEBackbone=Qwen2.5-7B-Instruct2025.05 | 0.27 | 0.06 | 0.14 | |
| MEMITBackbone=GPT-J2025.05 | 0.25 | 0.18 | 0.2 | |
| MEMITBackbone=GPT-J-6B2026.05 | 0.25 | 0.18 | 0.2 | |
| RECTBackbone=GPT-J2025.05 | 0.09 | 0.06 | 0.08 | |
| FTBackbone=Mistral-7B-v0.3, Number of Edits=20K2025.05 | 0.06 | 0.36 | 0.07 | |
| FTBackbone=Mistral-7B-v0.32026.05 | 0.06 | 0.36 | 0.07 | |
| AlphaEditBackbone=Mistral-7B-v0.3, Number of Edits=20K2025.05 | 0 | 0 | 0 | |
| MENDBackbone=Mistral-7B-v0.32025.05 | 0 | 0 | 0 | |
| MEMITBackbone=Mistral-7B-v0.32025.05 | 0 | 0 | 0 | |
| RECTBackbone=Mistral-7B-v0.32025.05 | 0 | 0 | 0 | |
| PRUNEBackbone=Mistral-7B-v0.32025.05 | 0 | 0 | 0 | |
| MENDBackbone=LLaMA-3-8B-Instruct2025.05 | 0 | 0 | 0 | |
| MEMITBackbone=Mistral-7B-v0.32026.05 | 0 | 0 | 0 | |
| ALPHAEDITBackbone=Mistral-7B-v0.32026.05 | 0 | 0 | 0 | |
| MENDBackbone=Mistral-7B-v0.32026.05 | 0 | 0 | 0 | |
| MENDBackbone=Llama-3-8B-Instruct2026.05 | 0 | 0 | 0 |