A multiple-choice test of everyday spatial understanding built from over 700 real-world photos, each paired with one verifiable question.
unassessed
| Category | multimodal |
|---|---|
| Subcategory | real-world spatial understanding |
| Page status | active |
| Metric | accuracy |
| Direction | higher_is_better |
| Unit | % |
| Dataset size | 765 |
| Dataset licence | CC BY-ND 4.0 |
| Publisher | xAI |
RealWorldQA shows a model a single photo taken from a vehicle or another everyday real-world setting and asks a multiple-choice question about what is in the scene, for example distances, positions, counts, or the relation between objects. The questions are meant to be easy for a person but still expose weaknesses in a model's basic spatial and physical-world understanding, rather than testing specialist knowledge or long chains of reasoning.
multiple-choice visual question answering over a single photo, 2-4 answer options per question, most with 3
| Model | Provider | Score | Card as of |
|---|---|---|---|
| GPT-4o | OpenAI | 75.4 | 2026-04 |
| GPT-4o (2024-05-13) | OpenAI | 75.4 | 2026-04 |
| GPT-4o (2024-08-06) | OpenAI | 75.4 | 2026-04 |
| GPT-4o (2024-11-20) | OpenAI | 75.4 | 2026-04 |
| NVLM D 72B | NVIDIA | 69.5 | 2026-04 |
| MiniCPM V 4 | OpenBMB | 68.5 | 2026-04 |
| MiniCPM V 4 5 | OpenBMB | 68.5 | 2026-04 |
| MiniCPM V 4 5 gguf | OpenBMB | 68.5 | 2026-04 |