Benchmarks · reasoning
ClockBench
Reading analog clocks. Trivial for people (90.7% average, 95% best) and hard for frontier models. Accuracy on the official leaderboard.
37 models · 37 results · data as of 2026-10-29 · clockbench.ai ↗
Leaderboard
Gemini 4 Argon high
All results
37 rows| # | Model | Effort | ClockBench |
|---|---|---|---|
| 1 | Gemini 4 Argon | high | 83.3% |
| 2 | Opus 5.5 | max | 77.2% |
| 3 | GPT-6 Astra | max | 75.0% |
| 4 | GPT-6.1 Sol | max | 74.4% |
| 5 | GPT-5.6 Sol | max | 68.3% |
| 6 | Gemini 3.8 Flash | max | 66.7% |
| 7 | Opus 5 | max | 65.5% |
| 8 | Fable 5.1 | max | 59.4% |
| 9 | Gemini 3.7 Flash | max | 53.3% |
| 10 | GPT-5.4 | high | 50.6% |
| 11 | GPT-5.5 | high | 46.1% |
| 12 | Muse Spark 1.3 | max | 42.2% |
| 13 | Qwen 3 | high | 39.4% |
| 14 | Fable 5 | max | 37.2% |
| 15 | Gemini 3.5 Flash | xhigh | 34.4% |
| 16 | Gemini 3.1 Pro | high | 32.2% |
| 17 | Gemini 3 Pro | high | 31.1% |
| 18 | Grok 4.5 | max | 21.7% |
| 19 | Grok 4.6 | max | 20.0% |
| 20 | Gemini 2.5 Pro | high | 18.9% |
| 21 | GPT-5.2 | high | 15.6% |
| 22 | Opus 4.7 | max | 15.0% |
| 23 | o3 pro | high | 14.4% |
| 24 | o3 | high | 12.2% |
| 25 | Gemini 2.5 Flash | medium | 11.1% |
| 26 | GPT-5 Pro | high | 11.1% |
| 27 | GPT-5 | high | 11.1% |
| 28 | Mistral Medium 3.1 | none | 10.6% |
| 29 | Opus 4.6 | max | 10.0% |
| 30 | Opus 4.1 | high | 8.9% |