1.google/gemma-4-31B-it
acc51.6%±2.1%
GPQA Extended
KnowledgeScienceReasoning
GPQA Extended includes the Main set plus additional questions that were authored but did not pass the rigorous filtering for the Main set.
Source paperLatest run: 2026-05-25
Benchmark results
Switch between the canonical ranking, release-date performance view, and score-size tradeoff.
28 models
2.google/Gemma-4-31B-IT-NVFP4
acc49.3%±2.1%
3.google/gemma-4-26B-A4B-it
acc44.3%±2.1%
4.Qwen/Qwen3-Next-80B-A3B-Instruct
acc40.8%±2.1%
5.google/gemma-3-27b-it
acc36.6%±2.1%
6.microsoft/phi-4
acc34.2%±2.0%
7.google/gemma-3-12b-it
acc33.3%±2.0%
8.Qwen/Qwen3-4B-AWQ
acc31.7%±2.0%
9.microsoft/phi-4-mini-instruct
acc31.3%±2.0%
10.Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8
acc28.9%±1.9%
11.openai/gpt-oss-120b
acc28.8%±1.9%
12.Qwen/Qwen3-32B
acc28.6%±1.9%
13.Qwen/Qwen3.6-35B-A3B
acc27.8%±1.9%
14.Qwen/Qwen3.5-35B-A3B
acc27.7%±1.9%
15.openai/gpt-oss-20b
acc27.7%±1.9%
16.Qwen/Qwen3-235B-A22B-Thinking-AWQ-2507
acc27.3%±1.9%
17.microsoft/phi-4-reasoning-plus
acc26.9%±1.9%
18.zai-org/GLM-4.5-Air-FP8
acc26.7%±1.9%
19.nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4
acc26.6%±1.9%
20.Qwen/Qwen3-14B
acc26.2%±1.9%
21.Qwen/Qwen3.5-122B-A10B-NVFP4
acc25.8%±1.9%
22.Qwen/Qwen3.6-27B
acc25.5%±1.9%
23.MiniMax/MiniMax-M2-AWQ
acc24.9%±1.9%
24.microsoft/phi-4-mini-reasoning
acc24.9%±1.9%
25.zai-org/GLM-4.5V-FP8
acc23.8%±1.8%
26.Qwen/Qwen3-8B
acc23.8%±1.8%
27.MiniMax/MiniMax-M2.1-AWQ
acc23.4%±1.8%
28.Qwen/Qwen3-4B
acc23.1%±1.8%
Caveats
Questions may be flawed, ambiguous, or trivial compared to the Main set.
How to cite
Citation
FrozeBench. "GPQA Extended." https://frozebench.com/benchmarks/gpqa-extended. Retrieved 2026-06-04.
BibTeX
@misc{frozebench_gpqa_extended,
title = {GPQA Extended},
howpublished = {\url{https://frozebench.com/benchmarks/gpqa-extended}},
year = {2026},
note = {FrozeBench. Retrieved 2026-06-04.}
}URL
https://frozebench.com/benchmarks/gpqa-extended