Benchmark
PIQA
PIQA (Physical Interaction: Question Answering) is a benchmark dataset for physical commonsense reasoning in natural language. It tests AI systems'…
- Modality
- text
- Categories
- physics, reasoning, general
- Openness
- unknown
- Source
- llm_stats
- Reported scores
- 11
Reported scores
llm_stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| ERNIE 4.5 | Baidu | 0.552 | 2025-06-25 | evidence |
| Gemma 2 27B | 0.832 | 2024-06-27 | evidence | |
| Gemma 2 9B | 0.817 | 2024-06-27 | evidence | |
| Gemma 3n E2B | 0.789 | 2025-06-26 | evidence | |
| Gemma 3n E2B Instructed LiteRT (Preview) | 0.789 | 2025-05-20 | evidence | |
| Gemma 3n E4B | 0.81 | 2025-06-26 | evidence | |
| Gemma 3n E4B Instructed LiteRT Preview | 0.81 | 2025-05-20 | evidence | |
| Hermes 3 70B | Nous Research | 0.8444 | 2024-08-15 | evidence |
| Phi 4 Mini | Microsoft | 0.776 | 2025-02-01 | evidence |
| Phi-3.5-mini-instruct | Microsoft | 0.81 | 2024-08-23 | evidence |
| Phi-3.5-MoE-instruct | Microsoft | 0.886 | 2024-08-23 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.