TriViewBench - Level 2: leaderboard
Metric: Accuracy (%) over the 3,447 Level-2 questions (low object count, high occlusion) of all three reasoning categories; 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: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Flash | 83.35 |
| 2 | Claude 3.7 Sonnet | 72.41 |
| 3 | Qwen 3 VL 32B Instruct | 69.39 |
| 4 | GPT-4o | 68.7 |
| 5 | Qwen 3 VL 8B Instruct | 64.93 |
| 6 | Qwen 3 VL 4B Instruct | 58.8 |
| 7 | Qwen 2.5 VL 32B Instruct | 49.38 |
| 8 | Qwen 2.5 VL 7B Instruct | 42.36 |
| 9 | InternVL3.5-8B | 40.91 |
Interactive version: theaggregate.ai/benchmark?slug=triviewbench-level-2 · How It Works · Data refreshed daily, snapshot 2026-09-29.