Scientific Reasoning on GPQA (pass@1)
70.2pass@1EvoEnv
Evaluation Results
| Method | Links | |
|---|---|---|
| EvoEnvBase Model=Qwen3-4B-Thinking-25072026.05 | 70.2 | |
| RLVEBase Model=Qwen3-4B-Thinking-25072026.05 | 67.7 | |
| R-ZeroBase Model=Qwen3-4B-Thinking-25072026.05 | 67.2 | |
| RLVEBase Model=Nemotron-Cascade-8B2026.05 | 66.2 | |
| EvoEnvBase Model=Nemotron-Cascade-8B2026.05 | 65.7 | |
| UntrainedBase Model=Qwen3-4B-Thinking-25072026.05 | 65.2 | |
| UntrainedBase Model=Nemotron-Cascade-8B2026.05 | 65.2 | |
| DAPOBase Model=Qwen3-4B-Thinking-25072026.05 | 64.6 | |
| DAPOBase Model=Nemotron-Cascade-8B2026.05 | 61.6 | |
| R-ZeroBase Model=Nemotron-Cascade-8B2026.05 | 58.1 | |
| RLVEBase Model=Qwen3-4B-Instruct-25072026.05 | 57.1 | |
| TTVSBackbone=Qwen3-8B2026.04 | 56.1 | |
| DAPOBase Model=Qwen3-4B-Instruct-25072026.05 | 56.1 | |
| EvoEnvBase Model=Qwen3-4B-Instruct-25072026.05 | 55 | |
| R-ZeroBase Model=Qwen3-4B-Instruct-25072026.05 | 54 | |
| TTRLBackbone=Qwen3-8B2026.04 | 53 | |
| UntrainedBase Model=Qwen3-4B-Instruct-25072026.05 | 52.5 | |
| TTVSBackbone=Qwen2.5-Instruct-7B2026.04 | 51.5 | |
| TTVSBackbone=Qwen3-4B2026.04 | 48.9 | |
| TTRLBackbone=Qwen2.5-Instruct-7B2026.04 | 48.5 | |
| Qwen3-8BTest-time method=Base2026.04 | 48 | |
| TTRLBackbone=Qwen3-4B2026.04 | 43 | |
| TTVSBackbone=LLaMA3.1-8B2026.04 | 37.3 | |
| TTVSBackbone=Qwen2.5-Instruct-3B2026.04 | 36.4 | |
| Qwen2.5-Instruct-7BTest-time method=Base2026.04 | 36.4 | |
| TTRLBackbone=LLaMA3.1-8B2026.04 | 34.1 | |
| TTRLBackbone=Qwen2.5-Instruct-3B2026.04 | 33.3 | |
| LLaMA3.1-8BTest-time method=Base2026.04 | 30.8 | |
| Qwen2.5-Instruct-3BTest-time method=Base2026.04 | 29.3 | |
| OpenReasoner-Zero-7BLabeled Data Size=129K2026.04 | 28.4 | |
| TTVSBackbone=Qwen2.5-Instruct-1.5B2026.04 | 28.3 | |
| SimpleRL-Zero-7BLabeled Data Size=8.9K2026.04 | 27.6 | |
| TTVSBackbone=Qwen3-1.7B-Instruct2026.04 | 26.3 | |
| Qwen3-4BTest-time method=Base2026.04 | 26.3 | |
| DeepSeek-R1-Distill-7BLabeled Data Size=800K2026.04 | 25.7 | |
| TTRLBackbone=Qwen2.5-Instruct-1.5B2026.04 | 25.3 | |
| TTRLBackbone=Qwen3-1.7B-Instruct2026.04 | 22.2 | |
| TTVSBackbone=LLaMA-3.2-3B-Instruct2026.04 | 17.7 | |
| Qwen2.5-Instruct-1.5BTest-time method=Base2026.04 | 16.7 | |
| TTRLBackbone=LLaMA-3.2-3B-Instruct2026.04 | 16.1 | |
| DeepSeek-R1-Distill-1.5BLabeled Data Size=800K2026.04 | 14.6 | |
| Qwen3-1.7B-InstructTest-time method=Base2026.04 | 13.1 | |
| LLaMA-3.2-3B-InstructTest-time method=Base2026.04 | 12.5 |