ChronoVision: leaderboard
ChronoVision (Seeing Time, 2026) tests whether vision-language models reason about when images were made, using Chinese dynasty artifacts, dated event photos and historical news. The overall score (0 to 100) is the mean of an artifact score (dynasty accuracy and chronological ordering), a shortcut score that discounts photo-ordering accuracy by how much grayscale changes it, and a news score (photo year accuracy discounted by its error, and image-text matching). Higher is better.
Metric: Overall score (0-100): mean of the Artifacts score (mean of dynasty accuracy and 50 x (1 + Kendall tau) ordering), the Shortcut score (color-pair accuracy x (1 - |grayscale accuracy gap|)) and the News score (mean of year accuracy discounted by MAE over 79 years and image-text matching accuracy); zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 67.17 | #266 |
| 2 | GPT-5.2 | 49.96 | #176 |
| 3 | Qwen 3 VL 235B A22B Instruct | 49.92 | #522 |
| 4 | Qwen 3 VL 8B Instruct | 44.47 | #981 |
| 5 | Qwen 3 VL 4B Instruct | 41.19 | #1279 |
| 6 | GLM-4.1V-9B (Thinking) | 37.35 | #1060 (GLM-4.1V-9B) |
| 7 | Qwen 2.5 VL 7B Instruct | 36.91 | #1833 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=chronovision · How It Works · Data refreshed daily, snapshot 2026-11-13.