Graduate-level Science Question Answering on GPQA
87Performance ScoreGPT-5
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-5Temperature=0, Access=Official API2026.01 | 87 | 1 | |
| Qwen2.5-72BTemperature=0, Access=Official API2026.01 | 72 | 2 | |
| Qwen2.5-32BTemperature=0, Access=Official API2026.01 | 70 | 3 | |
| Qwen-turboTemperature=0, Access=Official API2026.01 | 63 | 4 | |
| Llama3-8BTemperature=0, Hardware=2 4090 GPUs with 24GB memory2026.01 | 41 | 5 | |
| ZeroRL + ACTGUIDE-RLZero-shot=true, Chain-of-Thought (CoT)=true2026.05 | 36.93 | — | |
| ZeroRLZero-shot=true, Chain-of-Thought (CoT)=true2026.05 | 35.45 | — | |
| SFT + RLZero-shot=true, Chain-of-Thought (CoT)=true2026.05 | 29.85 | — | |
| SFT + ACTGUIDE-RLZero-shot=true, Chain-of-Thought (CoT)=true2026.05 | 29.57 | — | |
| SFTZero-shot=true, Chain-of-Thought (CoT)=true2026.05 | 29.15 | — | |
| DreamDecoding Strategy=Random, Diffusion Steps=2562026.02 | 27.9 | — | |
| CoDDDecoding Strategy=Entropy, Diffusion Steps=2562026.02 | 27.9 | — | |
| DreamDecoding Strategy=Low Confidence, Diffusion Steps=1282026.02 | 27.68 | — | |
| DreamDecoding Strategy=Low Confidence, Diffusion Steps=2562026.02 | 27.01 | — | |
| DreamDecoding Strategy=Low Confidence, Diffusion Steps=642026.02 | 26.12 | — | |
| DreamDecoding Strategy=Random, Diffusion Steps=642026.02 | 25 | — | |
| DreamDecoding Strategy=Random, Diffusion Steps=1282026.02 | 24.78 | — | |
| DreamDecoding Strategy=Margin, Diffusion Steps=2562026.02 | 24.33 | — | |
| DreamDecoding Strategy=Entropy, Diffusion Steps=2562026.02 | 24.33 | — | |
| CoDDDecoding Strategy=Entropy, Diffusion Steps=1282026.02 | 18.08 | — | |
| DreamDecoding Strategy=Margin, Diffusion Steps=1282026.02 | 14.06 | — | |
| DreamDecoding Strategy=Entropy, Diffusion Steps=1282026.02 | 14.06 | — | |
| CoDDDecoding Strategy=Entropy, Diffusion Steps=642026.02 | 10.94 | — | |
| DreamDecoding Strategy=Margin, Diffusion Steps=642026.02 | 8.04 | — | |
| DreamDecoding Strategy=Entropy, Diffusion Steps=642026.02 | 8.04 | — |