WorldAuditBench Spatial Consistency
Multimodal · 2026-09-30
The 51 WorldAuditBench tasks whose implanted anomaly belongs to the spatial consistency family: anomalies in appearance, visibility or lighting across viewpoints. A multimodal agent explores the scene with first-person views under a 40-action budget, using MCP tools to move, look, interact, recall earlier frames and report, and receives an anomaly-type hint and an in-context example; a GPT-6 Astra judge checks the report and its cited evidence against the task rubric. The score is the percentage of these tasks whose target anomaly is correctly identified.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 45.1 |
| Claude Opus 5 | 29.4 |
| Gemini 3.8 Flash | 29.4 |
| Muse Spark 1.3 | 13.7 |
| Qwen3.8 Flash | 3.9 |