CVT-Bench - Synthetic Competing NSC: leaderboard
Metric: Normalized survival coefficient (NSC): how long correct answers about the same tracked object pair persist across successive viewpoint probes, scaled so 0 is independent answering and 1 the maximum persistence the per-step accuracies allow (averaged over prefixes of 2 to 10 encounters; negative values retained), on the 100 CVT-Synthetic scenes (rendered CLEVR objects, 5,703 questions; relations left, right, front and behind), asked for the spatial relation between two tagged objects from the source view and from counterfactual viewpoints (yaw rotations and a top view), with the scene given as an image, 20 independent scenes interleaved in one request (CVT-Competing); higher is better. Source: arxiv.org. Saturation forecast: Around May 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 0.08 | #54 |
| 2 | Qwen 3.5 Plus (2026-02-15) | 0.02 | #114 |
| 3 | Kimi K2.5 | 0.02 | #139 |
| 4 | Qwen 3.5 397B A17B | 0.01 | #141 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=cvt-bench-synthetic-competing-nsc · How It Works · Data refreshed daily, snapshot 2026-10-11.