22.6%
#1 of 7
share of ClockQA clocks Gemini-2.0 read exactly right, the best of seven models; GPT-4o managed 8.1%
What the chart shows
Saxena et al. (2025) asked seven multimodal models to tell the time from 62 analogue clock images in six styles, including black dials and clocks without a second hand. Gemini-2.0 gave the exact time for 22.6% of them, the best result, ahead of GPT-4o (8.1%) and Claude-3.5-sonnet (6.5%).
The reasoning model GPT-o1 managed 4.8%, and Qwen2-VL-7B read none correctly. The task combines spotting the hands with simple arithmetic, so an error in one step carries into the next.
Key numbers
| Measure | Value |
|---|---|
| Gemini-2.0 | 22.58% |
| GPT-4o | 8.06% |
| Claude-3.5-sonnet | 6.45% |
| GPT-o1 | 4.84% |
| Qwen2-VL-7B | 0% |
How to read it
- Exact match means the predicted time matched the clock exactly; near misses count as wrong.
- A small test: with 62 images, each one is worth about 1.6 percentage points.
- Newer models were tested on a larger clock benchmark (related chart).
Source: Saxena et al., 2025, via the AI Index Report 2026 (Stanford HAI), figure 2.4.6: ClockQA: exact match. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion