SpaceVista: leaderboard
SpaceVista — spatial reasoning across scales from millimetres to kilometres in real scenes; overall accuracy.
Metric: Overall Accuracy (%). Source: peiwensun2000.github.io. Saturation forecast: Around 2029. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 33.8 | #152 |
| 2 | GPT-5 | 33.7 | #98 |
| 3 | InternVL3-78B | 33.5 | #352 |
| 4 | Claude Sonnet 4 | 29.7 | #199 |
| 5 | GPT-4o | 26.9 | #343 |
| 6 | InternVL3-38B | 26.5 | #405 |
| 7 | Claude Opus 4.1 | 26.4 | #140 |
| 8 | Gemini 2.5 Flash | 24.4 | #245 |
| 9 | GLM-4.5V | 23.3 | #346 |
| 10 | GLM-4.1V-9B (Thinking) | 23.1 | #472 (GLM-4.1V-9B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=spacevista · How It Works · Data refreshed daily, snapshot 2026-10-20.