Reasoning Performance and Sycophancy Evaluation on Reference Set N=500
90.5AccuracyRCA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RCAModel=GPT-5.1, Tier=Tier 4: RCA (Process Evaluation)2025.12 | 90.5 | 0 | |
| Self-ConsistencyModel=GPT-5.1, Tier=Tier 3: Self-Correction2025.12 | 86.1 | 8.2 | |
| ReflexionModel=GPT-5.1, Tier=Tier 3: Self-Correction2025.12 | 85.4 | 7.8 | |
| Self-RefineModel=GPT-5.1, Tier=Tier 3: Self-Correction2025.12 | 84.8 | 9.1 | |
| CoT-InstructedModel=GPT-5.1, Tier=Tier 3: Self-Correction2025.12 | 84.2 | 11.4 | |
| RCAModel=GPT-4o, Tier=Tier 4: RCA (Process Evaluation)2025.12 | 83.5 | 0 | |
| RCAModel=GPT-3.5, Tier=Tier 4: RCA (Process Evaluation)2025.12 | 74 | 0 | |
| DirectModel=GPT-4o, Tier=Tier 1: Direct Prompting2025.12 | 44.5 | 44 | |
| CoT-BalancedModel=GPT-4o, Tier=Tier 2: Chain-of-Thought2025.12 | 43 | 54.5 | |
| DirectModel=GPT-3.5, Tier=Tier 1: Direct Prompting2025.12 | 20.5 | 68 | |
| CoT-BalancedModel=GPT-3.5, Tier=Tier 2: Chain-of-Thought2025.12 | 6.5 | 87 |