TimeSpot - Daylight Phase: leaderboard
Metric: Accuracy (%) of the predicted daylight phase (such as sunrise, midday, sunset or night) on TimeSpot's 1,455 ground-level photographs from 80 countries, each answered zero-shot in a structured multi-field schema from the image alone; a Gemini-2.5-Flash judge normalizes each field (synonyms, abbreviations) and marks it correct or not; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 64.09 | #643 |
| 2 | O4 Mini | 51.79 | #172 |
| 3 | GLM-4.1V-9B (Thinking) | 47.76 | #457 (GLM-4.1V-9B) |
| 4 | GPT-5 Mini | 44.6 | #176 |
| 5 | Qwen 2.5 VL 32B Instruct | 44.54 | #443 |
| 6 | Llama 3.2 11B Instruct | 43.57 | #1112 |
| 7 | GLM-4.5V | 42.45 | #339 |
| 8 | Gemini 2.5 Flash (Non-reasoning) | 41.92 | #237 (Gemini 2.5 Flash) |
| 9 | Qwen 3 VL 235B A22B Instruct | 41.91 | #264 |
| 10 | GPT-5.2 | 41.66 | #105 |
| 11 | Qwen 2.5 VL 72B Instruct | 36.84 | #364 |
| 12 | Gemini 2.5 Flash (Thinking) | 36.56 | #237 (Gemini 2.5 Flash) |
| 13 | Mistral Medium 3.1 | 36.01 | #400 |
| 14 | InternVL3-78B | 34.91 | #345 |
| 15 | Llama 3.2 90B Vision Instruct | 33.88 | #611 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=timespot-daylight-phase · How It Works · Data refreshed daily, snapshot 2026-10-11.