General Knowledge Retention on General Utility
89.5Utility ScoreOriginal
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OriginalModel=Claude-Sonnet-42026.04 | 89.5 | — | 4.7 | |
| OriginalModel=GPT-4.12026.04 | 84.7 | — | 4.8 | |
| OriginalModel=GLM-42026.04 | 84.5 | — | 4 | |
| OriginalModel=DeepSeek-V32026.04 | 84.5 | — | 4.2 | |
| CAP(Ours)Model=Claude-Sonnet-42026.04 | 84.2 | — | 4.7 | |
| OriginalModel=Qwen3-235B-A22B2026.04 | 83.1 | — | 4.5 | |
| PromptingModel=Qwen3-235B-A22B2026.04 | 82.8 | — | 4.4 | |
| CAP(Ours)Model=DeepSeek-V32026.04 | 82.4 | — | 4.2 | |
| CAP(Ours)Model=GLM-42026.04 | 82.2 | — | 4 | |
| ICULModel=GPT-4.12026.04 | 81.5 | — | 4.6 | |
| ICULModel=Claude-Sonnet-42026.04 | 81.4 | — | 4.6 | |
| ICULModel=GLM-42026.04 | 81.3 | — | 3.9 | |
| CAP(Ours)Model=GPT-4.12026.04 | 80.6 | — | 4.8 | |
| CAP(Ours)Model=Qwen3-235B-A22B2026.04 | 80.2 | — | 4.5 | |
| PromptingModel=GPT-4.12026.04 | 80.1 | — | 4.7 | |
| ICULModel=Qwen3-235B-A22B2026.04 | 80.1 | — | 4.4 | |
| ICULModel=DeepSeek-V32026.04 | 79.9 | — | 4.1 | |
| PromptingModel=DeepSeek-V32026.04 | 79.8 | — | 4.2 | |
| OriginalModel=Qwen2.5-14B2026.04 | 79.5 | 1.1 | — | |
| PromptingModel=Claude-Sonnet-42026.04 | 79.2 | — | 4.7 | |
| PromptingModel=GLM-42026.04 | 76.7 | — | 3.9 | |
| LLMUModel=Qwen2.5-14B2026.04 | 75.9 | — | — | |
| CAP(Ours)Model=Qwen2.5-14B2026.04 | 75.9 | 1 | — | |
| NPOModel=Qwen2.5-14B2026.04 | 71.2 | — | — | |
| SPULModel=Qwen2.5-14B2026.04 | 70.7 | 1.5 | — | |
| ICULModel=Qwen2.5-14B2026.04 | 69.1 | 1.2 | — | |
| PromptingModel=Qwen2.5-14B2026.04 | 66.8 | 1.1 | — | |
| OriginalModel=Zephyr-7B2026.04 | 54.1 | 1.2 | — | |
| CAP(Ours)Model=Zephyr-7B2026.04 | 51.5 | 1.1 | — | |
| SPULModel=Zephyr-7B2026.04 | 50.6 | 1.5 | — | |
| NPOModel=Zephyr-7B2026.04 | 48.6 | — | — | |
| PromptingModel=Zephyr-7B2026.04 | 47.8 | 1.1 | — | |
| LLMUModel=Zephyr-7B2026.04 | 47.8 | — | — | |
| ICULModel=Zephyr-7B2026.04 | 44.5 | 1.3 | — |