Our new X account is live! Follow @wizwand_team for updates

Question Answering on GPQA (test)

45.5Accuracy

UPA

Updated 4d ago

Evaluation Results

Method
UPA 2026.01	45.5	-	-	-	-
kNN-MoE 2026.01	45.45	-	-	-	-
kNN-MoE 2026.01	44.95	-	-	-	-
5-shot 2026.01	44.44	-	-	-	-
Zero-shot 2026.01	43.94	-	-	-	-
SFT (Router Only) 2026.01	43.94	-	-	-	-
OPRO 2026.01	43.3	-	-	-	-
Step-Back 2026.01	42.4	-	-	-	-
Qwen3 8B 2026.02	41.96	-	-	-	-
SFT 2026.01	41.92	-	-	-	-
SPO 2026.01	41.8	-	-	-	-
CoT 2026.01	41.6	-	-	-	-
SFT (Router Only) 2026.01	41.41	-	-	-	-
Zero-shot 2026.01	41.41	-	-	-	-
PromptAgent 2026.01	41.3	-	-	-	-
APE 2026.01	41.1	-	-	-	-
PromptBreeder 2026.01	40.9	-	-	-	-
QWEN3-14B 2026.01	40.34	-	-	-	-
RaR 2026.01	40.2	-	-	-	-
TextGrad 2026.01	40.2	-	-	-	-
SFT 2026.01	39.9	-	-	-	-
IO 2026.01	38.9	-	-	-	-
Llama3.1 7B 2026.02	37.71	-	-	-	-
MCDIFFUSE 2026.02	37.19	-	-	-	-
5-shot 2026.01	36.87	-	-	-	-
QWEN3-4B 2026.01	36.64	-	-	-	-
ICL_bio 2024.04	36.2	-	-	-	-
Qwen2.5 7B 2026.02	35.94	-	-	-	-
ICL_same 2024.04	35.8	-	-	-	-
DEEPSEEK-R1-QWEN-8B 2026.01	35.18	-	-	-	-
N/A 2024.04	34.4	-	-	-	-
Random_same 2024.04	33.7	-	-	-	-
Random_diff 2024.04	33.1	-	-	-	-
Random_bio 2024.04	32.6	-	-	-	-
base 2025.01	32.11	-	-	-	-
GPT-4 2025.01	32.1	-	-	-	-
TAMA 2025.01	31.92	-	-	-	-
MISTRAL-7B 2026.01	31.8	-	-	-	-
Relevant 2024.04	31.6	-	-	-	-
LLAMA-3.1-8B 2026.01	31.38	-	-	-	-
DART 2026.01	30.97	-	-	-	-
DART 2026.01	30.59	-	-	-	-
GPT-3.5 2025.01	29.8	-	-	-	-
kNN-MoE 2026.01	29.8	-	-	-	-
LLAMA-3.2-3B 2026.01	29.63	-	-	-	-
Zero-shot 2026.01	27.27	-	-	-	-
DART 2026.01	27	-	-	-	-
DART 2026.01	26.16	-	-	-	-
DART 2026.01	25.9	-	-	-	-
DEEPSEEK-R1-LLAMA-8B 2026.01	25.57	-	-	-	-
SFT (Router Only) 2026.01	24.24	-	-	-	-
DART 2026.01	24.13	-	-	-	-
DART 2026.01	22.77	-	-	-	-
5-shot 2026.01	21.72	-	-	-	-
SFT 2026.01	21.72	-	-	-	-
QwQ-32B 2025.02	-	75.6	39.8	68.4	58.1
Llama3.3-70B 2025.02	-	54.7	31.2	52.6	43.4
DeepSeek-R1 2025.02	-	86.8	56.1	63.8	71.5
GPT-4o 2025.02	-	59.5	40.2	61.6	50
o1 2025.02	-	92.8	64.7	69.2	78
o3-mini-low 2025.02	-	-	-	-	70.6
o3-mini-mid 2025.02	-	-	-	-	76.8
o3-mini-high 2025.02	-	-	-	-	79.7
RAgent w/QwQ-32B 2025.02	-	76.7	46.2	68.4	61.6
RAgent w/DeepSeek-R1 2025.02	-	87.7	58.2	65.7	72.9
SearchO1 w/QwQ-32B 2025.02	-	77.9	47.3	78.9	63.6
SearchO1 w/DeepSeek-R1 2025.02	-	90.2	61.3	71.4	74.6
Agentic Reasoning w/QwQ-32B 2025.02	-	88.1	58.3	79.6	69.7
Agentic Reasoning w/DeepSeek-R1 2025.02	-	94.5	73.7	80.5	81.2
Qwen2.5-32B 2025.01	-	57	33.3	52.6	45.5
Qwen2.5-Coder-32B 2025.01	-	37.2	25.8	57.9	33.8
QwQ-32B 2025.01	-	75.6	39.8	68.4	58.1
Qwen2.5-72B 2025.01	-	57	37.6	68.4	49
Llama3.3-70B 2025.01	-	54.7	31.2	52.6	43.4
DeepSeek-R1-Lite 2025.01	-	-	-	-	58.5
GPT-4o 2025.01	-	59.5	40.2	61.6	50.6
o1-preview 2025.01	-	89.4	59.9	65.9	73.3
RAG-Qwen2.5-32B 2025.01	-	57	37.6	52.6	47.5
RAG-QwQ-32B 2025.01	-	76.7	38.7	73.7	58.6
RAgent-Qwen2.5-32B 2025.01	-	58.1	33.3	63.2	47
RAgent-QwQ-32B 2025.01	-	76.7	46.2	68.4	61.6
Search-o1 2025.01	-	77.9	47.3	78.9	63.6
Base 2026.01	-	-	-	-	33.3
Low-temperature 2026.01	-	-	-	-	30.3
Best-of-N 2026.01	-	-	-	-	29.7
MCMC Power Sampling 2026.01	-	-	-	-	34.9
Power Sampling 2026.01	-	-	-	-	36.4