MultiView-Bench - 3D DoF 2: leaderboard
Given six rendered views with visible world axes, state a target object position relative to a central object when both lie in a shared plane; from MultiView-Bench (2026). Accuracy in percent, higher is better.
Metric: Accuracy (%; six views, objects share the central object plane; fraction of 100 procedurally generated Blender scenes whose relative position of a target object is answered exactly as (+/-X or 0, +/-Y or 0, +/-Z or 0) in a visible world coordinate frame; mean over three runs where printed (Claude 3.7 Sonnet one run); x100). Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Medium) | 51.3 | #30 (GPT-5.6 Sol) |
| 2 | GPT-5.6 Luna (Medium) | 44.3 | #121 (GPT-5.6 Luna) |
| 3 | GPT-5 | 37.3 | #167 |
| 4 | GPT-5.6 Terra (Medium) | 37 | #62 (GPT-5.6 Terra) |
| 5 | O3 | 36.8 | #228 |
| 6 | Gemini 3.1 Pro (Preview) (Medium) | 33.7 | #94 (Gemini 3.1 Pro (Preview)) |
| 7 | Gemini 3.5 Flash (Medium) | 21 | #101 (Gemini 3.5 Flash) |
| 8 | Claude 3.7 Sonnet | 15 | #458 |
| 9 | Gemini 2.5 Pro | 12.2 | #266 |
| 10 | Gemini 2.5 Flash | 12 | #453 |
| 11 | Claude Haiku 4.5 (Thinking) | 11 | #568 (Claude Haiku 4.5) |
| 12 | GPT-4o | 7.3 | #741 |
| 13 | Claude Sonnet 4 | 7.2 | #400 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=multiview-bench-3d-dof-2 · How It Works · Data refreshed daily, snapshot 2026-11-13.