Graduate-Level Reasoning on GPQA (Pass@1)
70.18Pass@1Phi-4-mini + Mistral3-3B
Evaluation results
| Method | Links | |
|---|---|---|
| Phi-4-mini + Mistral3-3BTraining=CORE2026.01 | 70.18 | |
| Phi-4-mini + Mistral3-3B + OracleTraining=SD-E²2026.01 | 48.12 | |
| EGPSGBackbone=Qwen2.5-Math-7B, Variant=MTM2026.06 | 42.4 | |
| Scalable PSBackbone=Qwen2.5-Math-7B2026.06 | 40.9 | |
| GRPO (MATH)Backbone=Qwen2.5-Math-7B, Note=RL upper bound reference2026.06 | 39.9 | |
| MH Power SamplingBackbone=Qwen2.5-Math-7B2026.06 | 38.9 | |
| EGPSLBackbone=Qwen2.5-Math-7B, Variant=MTM2026.06 | 38.4 | |
| MH Power SamplingBackbone=Phi-3.5-mini-instruct2026.06 | 36.4 | |
| EGPSLBackbone=Qwen2.5-7B, Variant=MTM2026.06 | 35.9 | |
| GRPO (MATH)Backbone=Phi-3.5-mini-instruct, Note=RL upper bound reference2026.06 | 35.9 | |
| APPS (p-only)Backbone=Qwen2.5-7B2026.06 | 35.4 | |
| EGPSGBackbone=Qwen2.5-7B, Variant=MTM2026.06 | 35.4 | |
| GRPO (MATH)Backbone=Qwen2.5-7B, Note=RL upper bound reference2026.06 | 35.4 | |
| Low-tempBackbone=Qwen2.5-Math-7B2026.06 | 35.3 | |
| Power-SMCBackbone=Qwen2.5-Math-7B2026.06 | 34.9 | |
| Scalable PSBackbone=Qwen2.5-7B2026.06 | 34.9 | |
| Best-of-NBackbone=Qwen2.5-Math-7B, N=322026.06 | 34.3 | |
| EGPSGBackbone=Phi-3.5-mini-instruct, Variant=MTM2026.06 | 34.3 | |
| EGPSLBackbone=Phi-3.5-mini-instruct, Variant=MTM2026.06 | 33.8 | |
| Ministral-3-8B-ReasoningTraining=SD-E²2026.01 | 33 | |
| APPS (p-only)Backbone=Qwen2.5-Math-7B2026.06 | 32.3 | |
| MH Power SamplingBackbone=Qwen2.5-7B2026.06 | 31.8 | |
| Power-SMCBackbone=Qwen2.5-7B2026.06 | 31.3 | |
| Low-tempBackbone=Qwen2.5-7B2026.06 | 30.3 | |
| Qwen2.5-7B-InstructTraining=SD-E²2026.01 | 30 | |
| Low-tempBackbone=Phi-3.5-mini-instruct2026.06 | 29.3 | |
| Best-of-NBackbone=Qwen2.5-7B, N=322026.06 | 28.2 | |
| BaseBackbone=Qwen2.5-Math-7B2026.06 | 27.8 | |
| BaseBackbone=Qwen2.5-7B2026.06 | 27.8 | |
| BaseBackbone=Phi-3.5-mini-instruct2026.06 | 27.3 | |
| Phi-3-small-8k-InstructTraining=SD-E²2026.01 | 26 |