General Reasoning on MMLU-P
75.6AccuracyInstruct
Evaluation results
| Method | Links | ||
|---|---|---|---|
| InstructBackbone=Qwen3-14B2026.01 | 75.6 | — | |
| ReasoningGuardBackbone=Qwen3-14B2026.01 | 75.6 | — | |
| Self-ReminderBackbone=Qwen3-14B2026.01 | 74.2 | — | |
| Self-GuardBackbone=Qwen3-14B2026.01 | 74.2 | — | |
| ReasoningGuardBackbone=Qwen3-8B2026.01 | 73.8 | — | |
| Self-ReminderBackbone=Qwen3-8B2026.01 | 72.4 | — | |
| InstructBackbone=Qwen3-8B2026.01 | 72.2 | — | |
| SafeKeyBackbone=Qwen3-8B2026.01 | 71.4 | — | |
| SafeKeyBackbone=Qwen3-14B2026.01 | 71.4 | — | |
| Alpha-steerBackbone=Qwen3-14B2026.01 | 71.2 | — | |
| Self-GuardBackbone=Qwen3-8B2026.01 | 70.8 | — | |
| STAR-1Backbone=Qwen3-14B2026.01 | 70.8 | — | |
| STAR-1Backbone=Qwen3-8B2026.01 | 68.6 | — | |
| SafeChainBackbone=Qwen3-14B2026.01 | 68.2 | — | |
| Self-ReminderBackbone=Qwen3-4B2026.01 | 67.8 | — | |
| ReasoningGuardBackbone=Qwen3-4B2026.01 | 67 | — | |
| InstructBackbone=Qwen3-4B2026.01 | 65.8 | — | |
| Self-GuardBackbone=Qwen3-4B2026.01 | 65.6 | — | |
| Alpha-steerBackbone=Qwen3-8B2026.01 | 63.2 | — | |
| SafeChainBackbone=Qwen3-8B2026.01 | 63 | — | |
| STAR-1Backbone=Qwen3-4B2026.01 | 61.8 | — | |
| Alpha-steerBackbone=Qwen3-4B2026.01 | 60.8 | — | |
| SafeChainBackbone=Qwen3-4B2026.01 | 56.8 | — | |
| SafeKeyBackbone=Qwen3-4B2026.01 | 55 | — | |
| BaseBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.05 | — | 49 | |
| BaseBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.05 | — | 62 | |
| DRIFT-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Ours, Max turns=52026.05 | — | 57.2 | |
| DRIFT-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Ours, Max turns=52026.05 | — | 62.6 | |
| PPOBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.05 | — | 49.7 | |
| PPOBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.05 | — | 60 | |
| SCoRe-2turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=22026.05 | — | 55.2 | |
| SCoRe-2turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=22026.05 | — | 62.1 | |
| SFTBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.05 | — | 50.4 | |
| SFTBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.05 | — | 58.9 | |
| SFT-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=52026.05 | — | 53.3 | |
| SFT-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=52026.05 | — | 57.4 | |
| STaR-2turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=22026.05 | — | 57.1 | |
| STaR-2turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=22026.05 | — | 62.3 | |
| UFO-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=52026.05 | — | 57.1 | |
| UFO-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=52026.05 | — | 64.3 |