# Gemini 3.1 Pro Preview prompt sheet

Model ID: `gemini-3.1-pro-preview`.

## Submitted task text

```text
You are watching 0.8 seconds of video from a camera mounted on a robot gripper. Answer in JSON only, with exactly two keys. contact: yes if the gripper fingers are touching an object or a surface at any point in this clip, no if they are not, unsure if you cannot tell. onset_ms: if contact begins during this clip, the millisecond from the start of the clip at which it begins (an integer from 0 to 800); otherwise null.
```

## Input and request settings

| Setting | Recorded value |
| --- | --- |
| Source window | 800ms, wrist RGB, 256x256 |
| Sampling | Nearest source frame to the 24Hz grid within the half-open source window; 20 selections |
| Submitted media | Inline H.264 MP4, CRF 23, yuv420p, no audio |
| Time information | Uniform 24fps presentation clock; explicit video metadata fps=24 |
| Model media setting | LOW media resolution |
| Task-message order | Video, then task text |
| Temperature | Explicit 0 |
| Reasoning setting | Not explicitly set in submitted generation config |
| Output mode | API JSON response mode |
| Output token cap | Not explicitly set in submitted generation config |
| Submission mode | Provider batch API |
| Tools/history | None |

The prompt asks for visible contact. The loading-onset reference marks a force change.

Twenty selected frames use a uniform 24 fps presentation clock. Their nominal encoded duration is about 0.833 seconds; source-frame selection and the requested timestamp range use the 800 ms source window.