MultihopSpatial - Acc@50IoU: leaderboard
Metric: Acc@50IoU (%): a prediction counts only when the answer is right and the predicted box overlaps the target at IoU 0.5 or more, over all 4,500 MultihopSpatial questions (1-3 hops, ego- and exo-centric), one prompt template, answer and target bounding box required; higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 40.6 | #77 |
| 2 | Gemini 3 Flash | 40.2 | #93 |
| 3 | Qwen 3 VL 32B (Thinking) | 37.4 | #287 (Qwen 3 VL 32B) |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 36.3 | #228 (Qwen 3 VL 235B A22B) |
| 5 | GLM-4.6V (Non-reasoning) | 35.2 | #309 (GLM-4.6V) |
| 6 | Qwen 3 VL 235B A22B Instruct | 34.8 | #264 |
| 7 | GLM-4.6V (Thinking) | 34.7 | #309 (GLM-4.6V) |
| 8 | Qwen 3 VL 32B Instruct | 33.4 | #276 |
| 9 | Qwen 3 VL 8B Instruct | 31.3 | #401 |
| 10 | Qwen 3 VL 4B Instruct | 31 | #506 |
| 11 | Qwen 3 VL 8B (Thinking) | 29.5 | |
| 12 | Qwen 3 VL 4B (Thinking) | 28.7 | #471 (Qwen 3 VL 4B) |
| 13 | GPT-5.2 (xHigh) | 11.5 | #105 (GPT-5.2) |
| 14 | Claude Opus 4.5 (Thinking) | 4.7 | #79 (Claude Opus 4.5) |
| 15 | Claude Sonnet 4.5 (Thinking) | 4.3 | #138 (Claude Sonnet 4.5) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=multihopspatial-acc-50iou · How It Works · Data refreshed daily, snapshot 2026-10-11.