MMSI-Bench: leaderboard
Multi-image spatial intelligence benchmark testing VLMs on positional relationships, attribute reasoning, and motion reasoning across multiple images.
Metric: Accuracy (%). Source: github.com. Saturation forecast: Around 2030. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Median Human | 97.2 | |
| 2 | Gemini 3 Pro | 49.2 | #80 |
| 3 | GPT-5 | 41.9 | #98 |
| 4 | O3 | 41 | #124 |
| 5 | GPT-4.5 | 40.3 | #274 |
| 6 | Gemini 2.5 Pro (Thinking) | 37 | #152 (Gemini 2.5 Pro) |
| 7 | Gemini 2.5 Pro | 36.9 | #152 |
| 8 | Doubao-1.5-Pro | 33 | #287 |
| 9 | GPT-4.1 | 30.9 | #248 |
| 10 | GPT-4o | 30.3 | #343 |
| 11 | Claude 3.7 Sonnet (Thinking) | 30.2 | #249 (Claude 3.7 Sonnet) |
| 12 | InternVL2.5-2B | 29 | #1249 |
| 13 | InternVL3-78B | 28.5 | #352 |
| 14 | InternVL2.5-78B | 28.5 | #431 |
| 15 | Qwen 2.5 VL 7B | 25.9 | #643 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmsi-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.