ChronoVision - Event Photo Ordering (Grayscale Conflict): leaderboard
Metric: Accuracy (%) on the 1,200 SPEED photo pairs with the later photo converted to grayscale, so that the grayscale-means-old shortcut points to the wrong answer; zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 78.75 |
| 2 | GPT-5.2 | 56.83 |
| 3 | Qwen 3 VL 8B Instruct | 56.75 |
| 4 | Qwen 3 VL 235B A22B Instruct | 55.75 |
| 5 | Qwen 3 VL 4B Instruct | 47.58 |
| 6 | Qwen 2.5 VL 7B Instruct | 40.83 |
| 7 | GLM-4.1V-9B (Thinking) | 32.08 |
| 8 | InternVL3.5-8B | 24.92 |
Interactive version: theaggregate.ai/benchmark?slug=chronovision-event-photo-ordering-grayscale-conflict · How It Works · Data refreshed daily, snapshot 2026-09-29.