// BENCHMARK

PIQA benchmark

AI model leaderboard for the PIQA benchmark. Compare how large language models score on PIQA, see the full ranking, and understand what this AI benchmark measures. GPT-4o mini currently leads with 88.7. Physical Interaction QA — commonsense benchmark on physical-world reasoning.

Leaderboard

# Model Organization Score Variant Source
#1 GPT-4o mini OpenAI 88.7 cited-phi35moe official ↗
#2 Phi-3.5-MoE Microsoft 88.6 instruct-5-shot official ↗
#3 Gemini 1.5 Flash (Sep 2024) Google DeepMind 87.5 cited-phi35moe official ↗
#4 Gemma 2 9B Google DeepMind 83.7 instruct-cited-phi35moe official ↗
#5 Mistral NeMo Mistral AI 83.5 instruct-cited-phi35moe official ↗
#6 DCLM 7B Apple 81.28 from-hf-readme official ↗
#7 Llama 3.1-8B Meta AI 81.2 instruct-cited-phi35moe official ↗
#8 Llama 3.1-8B Meta AI 78.9 instruct-0-shot-cited-falcon3 official ↗
#9 Falcon3-7B Technology Innovation Institute 78.8 instruct-0-shot official ↗
#10 phi-3.5-mini Microsoft 78.2 cited-phi4-mini official ↗
#11 Phi-4 Mini Microsoft 77.6 instruct-5-shot official ↗
#12 Qwen2.5-7B Qwen 76.2 cited-phi4-mini official ↗
#13 Qwen2.5-7B Qwen 73.7 instruct-0-shot-cited-falcon3 official ↗
#14 Ministral 8B Mistral AI 73.2 cited-phi4-mini official ↗
#15 Qwen2.5-3B Qwen 72.6 cited-phi4-mini official ↗
#16 Llama 3.2 3B Meta AI 68.2 cited-phi4-mini official ↗
#17 Llama 3.1-8B Meta AI 45.8 base-cited-marin official ↗
#18 Marin 8B Marin 44.2 base-from-hf-readme official ↗

Frequently asked questions about PIQA

What is the PIQA benchmark?

Physical Interaction QA — commonsense benchmark on physical-world reasoning.

How is the PIQA benchmark scored?

PIQA is scored using the accuracy metric, where a higher score is better. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.

Which AI model scores highest on PIQA?

As of the latest reported scores on GenAIList, GPT-4o mini achieves the highest result on PIQA with a score of 88.7.

Is a higher PIQA score better?

Yes. On PIQA a higher score indicates better performance, so models near the top of the leaderboard are the strongest.