Instruction Following on Human Evaluation Set 200 instruction-response pairs (test)
4.38Constraint ScoreClaude + CRGC
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Claude + CRGCBase Model=Claude, Prompting Strategy=Constraint Relationship Graph Completion (CRGC)2026.06 | 4.38 | 4.42 | 4.4 | |
| GPT-4o + CRGCBase Model=GPT-4o, Prompting Strategy=Constraint Relationship Graph Completion (CRGC)2026.06 | 4.29 | 4.37 | 4.33 | |
| Claude + T2Base Model=Claude, Prompting Strategy=T22026.06 | 3.75 | 4.35 | 4.05 | |
| Claude + CoTBase Model=Claude, Prompting Strategy=Chain-of-Thought (CoT)2026.06 | 3.69 | 4.31 | 4 | |
| GPT-4o + T2Base Model=GPT-4o, Prompting Strategy=T22026.06 | 3.64 | 4.29 | 3.97 | |
| GPT-4o + CoTBase Model=GPT-4o, Prompting Strategy=Chain-of-Thought (CoT)2026.06 | 3.57 | 4.23 | 3.9 | |
| Claude + SPBase Model=Claude, Prompting Strategy=Standard Prompting (SP)2026.06 | 3.35 | 4.22 | 3.79 | |
| GPT-4o + SPBase Model=GPT-4o, Prompting Strategy=Standard Prompting (SP)2026.06 | 3.21 | 4.15 | 3.68 |