SceneActBench - Camera: leaderboard

Metric: Task score (0-100): 100 furnished 3D-FRONT scenes: recover the 6-DoF camera pose of one view with known field of view, scored by position and angle error against 4 m and 90 degree references; every configuration drives one headless Blender through the same MCP tool interface (scene and object inspection, Python execution, rendering) with task step budgets; each native geometric error or F-score is mapped to a 0 to 100 case score against a fixed reference, invalid outputs score 0, cases are averaged per task. Source: arxiv.org. Saturation forecast: Around September 2028. 11 models tracked.

Top models

#ModelScore
1Seed 2.0 Pro (High)34.5
2Claude Opus 4.6 (High)34
3Gemini 3.1 Pro (Preview) (High)33.9
4GPT-5.4 (Medium)29.6
5Claude Sonnet 5 (High)27.3
6GPT-5.4 (High)26.4
7MiniMax M3 (High)25.6
8Kimi K2.6 (Thinking)24.6
9Step 3.7 Flash (High)13.2

Interactive version: theaggregate.ai/benchmark?slug=sceneactbench-camera · How It Works · Data refreshed daily, snapshot 2026-09-29.