ClockBench: leaderboard

Visual reasoning benchmark testing if AI can read analog clocks. 180 clocks, 720 questions covering time reading, arithmetic, rotation, and timezone conversion. Human baseline: 90.7%.

Metric: Accuracy (%). Source: clockbench.ai. Status: years away from saturation. 35 models tracked.

Top models

#ModelScore
1Best Expert93.7
2Median Human90.7
3GPT-5.6 Sol (Max)66.7
4Gemini 3.8 Flash62.8
5Claude Opus 5 (Max)60.7
6Claude Fable 5.1 (Max)53.3
7GPT-5.4 (High)50.6
8Gemini 3.7 Flash48.3
9GPT-5.5 (High)46.1
10Muse Spark 1.342.2
11Qwen 3 VL 235B A22B Instruct39.4
12Claude Fable 535
13Gemini 3.1 Pro (Preview)32.2
14Gemini 3.5 Flash31.1
15Gemini 3 Pro28.9

Interactive version: theaggregate.ai/benchmark?slug=clockbench · How It Works · Data refreshed daily, snapshot 2026-09-05.