TriViewBench - Local Decision: leaderboard
TriViewBench (2026) tests multi-view structural reasoning in multimodal models on synthetic 3D tower scenes shown as front, side and top views, with controlled object count and occlusion; models answer directly within 128 tokens. This board is accuracy in percent on the 3,508 Local Decision questions, which compare tower heights and relative object positions, across all complexity levels; higher is better.
Metric: Accuracy (%) over the 3,508 Local Decision questions (tallest tower and relative x, y, z position of two objects), all four complexity levels; three orthographic views (front, side, top) of synthetic 3D tower scenes, direct prompting with a concise answer within 128 tokens, greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 89.65 | #453 |
| 2 | Qwen 3 VL 32B Instruct | 86.83 | #585 |
| 3 | Qwen 3 VL 8B Instruct | 84.49 | #981 |
| 4 | Claude 3.7 Sonnet | 80.99 | #458 |
| 5 | GPT-4o | 74.52 | #741 |
| 6 | Qwen 3 VL 4B Instruct | 70.67 | #1279 |
| 7 | Qwen 2.5 VL 32B Instruct | 60.83 | #1053 |
| 8 | Qwen 2.5 VL 7B Instruct | 54.1 | #1833 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=triviewbench-local-decision · How It Works · Data refreshed daily, snapshot 2026-11-13.