Back to Devoured - September 21, 2026
DEVOURED
Segment Anything Model (SAM) 3.1

Segment Anything Model (SAM) 3.1

AI computer-vision Meta
Meta released SAM 3.1, a model capable of real-time object segmentation and tracking in video and images via its API.
What: SAM 3.1 is available on the Meta Model API, priced at $2.50 per 1,000 images and $0.20 per 1,000 video frames, enabling zero-shot segmentation without fine-tuning.
Why it matters: Meta is increasingly productizing its perception models through a unified API, shifting from pure research releases to production-grade services that directly compete with specialized computer vision providers.
Takeaway: Test the model by pointing an OpenAI SDK-compatible client at the Meta Model API.
Decoder
  • SAM (Segment Anything Model): A computer vision architecture developed by Meta that can identify, segment, and mask objects in an image based on pixel-level data.
Original article

Segment Anything Model (SAM) 3.1

Detect, segment and track objects in images and video on Meta Model API, with nothing to host or tune.

Meet SAM 3.1

SAM 3.1 architecture

More ways to build

Models and pricing

Leading perception with purpose-built inference

SAM 3.1 is our leading perception model for object detection, segmentation and tracking, served on inference built specifically for its DETR architecture. Get production throughput on day one, with nothing to stand up or tune.

Integrated with other Meta models

SAM 3.1 shares its envelope, keys, docs and billing with every other Meta model, so segmentation and video tracking drop smoothly into the pipeline you already have. Reason and generate with Muse Spark, turn speech into text with Muse Voice Transcribe, and detect, segment and track objects with SAM 3.1, all through Meta Model API.

Detection, segmentation and tracking in one model

Most multimodal models give you labels and bounding boxes and leave the rest to you. SAM 3.1 detects the objects, returns pixel-precise segmentation masks and tracks them through video with identity preserved, all from a single call. Zero-shot, with no training data, fine-tuning or extra models required.

Models and pricing

  • sam-3-1 Segmentation: $2.50/1k images, $0.20/1k frames

Cookbooks and quickstarts

Quickstart: Your first working request in under five minutes. Point your existing OpenAI SDK compatible client at Meta Model API.

SAM Overview: Use text prompts to identify, segment, and follow any object in images or video.

Segmenting: Give SAM 3.1 a short noun phrase naming one thing, and it returns a box and a mask for every match.

Explore all 67 articles in today's digest

Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi