ClockBench Accuracy
Multimodal · 2025-09-02
ClockBench Accuracy is the public leaderboard's primary reading metric across 180 analog clocks, including 37 invalid clocks. The grader accepts a correctly identified invalid clock without checking time fields, so an always-invalid answer earns 37/180; this mixed validity and open-answer task has no curated random-guessing floor.
Top models (higher is better)
| Model | Score |
|---|---|
| Top Human Performance (ClockBench) | 95.0 |
| Human Baseline (ClockBench) | 90.7 |
| Gemini 4 Argon | 83.3 |
| Claude Opus 5.5 | 77.2 |
| GPT-6 Astra | 75.0 |
| GPT-6.1 Sol | 74.4 |
| GPT-5.6 Sol | 68.3 |
| Gemini 3.8 Flash | 66.7 |
| Claude Opus 5 | 65.5 |
| Claude Fable 5.1 | 59.4 |
| Gemini 3.7 Flash | 53.3 |
| GPT-5.4 | 50.6 |
| GPT-5.5 | 46.1 |
| Muse Spark 1.3 | 42.2 |
| Qwen3-VL-235B-A22B-Instruct | 39.4 |
| Claude Fable 5 | 37.2 |
| Gemini 3.5 Flash | 34.4 |
| Gemini 3.1 Pro Preview | 32.2 |
| Gemini 3 Pro Preview | 31.1 |
| Grok 4.5 | 21.7 |
| Grok 4.6 | 20.0 |
| Gemini 2.5 Pro | 18.9 |
| GPT-5.2 | 15.6 |
| Gemini Robotics-ER 1.5 | 15.0 |
| Opus 4.7 | 15.0 |
| Qwen3 VL 235B A22B Thinking | 14.4 |
| o3-pro | 14.4 |
| o3 | 12.2 |
| Gemini 2.5 Flash | 11.1 |
| GPT-5 | 11.1 |
| GPT-5 Pro | 11.1 |
| Mistral Medium 3.1 | 10.6 |
| Opus 4.6 | 10.0 |
| GPT-5 Mini | 8.9 |
| Opus 4.1 | 8.9 |
| Sonnet 4.5 | 7.8 |
| Sonnet 4 | 6.7 |
| GPT-4o (source unspecified) | 6.7 |
| Qwen2.5-VL-72B-Instruct | 6.1 |
| GPT-5 Nano | 3.9 |