09 · Bonus: images · 6 / 6How long, and what to know
One photo, three questions, one call
0.5–0.8 susually
dog · nose
446 ms
dog · main subject
484 ms
dog · eye
728 ms
pug · main subject
844 ms
cars · once
2380 ms
Measured in our demo, through Vercel AI Gateway, October 2026.
Network included: try it again before you rely on it.
Before you use it
🧪
Experimental
Opt in with TypeSafe-Danger-Zone-Images.
Not recommended for production yet.
🖼️
PNG or JPEG only
WebP and GIF are rejected. Rotate the pixels upright.
📐
280–1568 px per side
Resize phone photos first. Up to 16 MiB per image.
💰
Billed as input tokens
Same price as text. Output is still free.
Fast enough for a UI. Still experimental.