SAM 3.1 in the Meta Model API

Meta's object detection and tracking model is now a hosted API call instead of something you run yourself.

Detects objects from a short text phrase, returns pixel-precise segmentation masks, and tracks them across video frames with identity preserved, all from one call
Runs on inference built specifically for SAM 3.1's DETR architecture, so there is nothing to host or tune
Priced at $2.50 per 1,000 images and $0.20 per 1,000 video frames
Comes with a playground for trying your own media, and works with clients built for the OpenAI SDK