Skip to main content

Meta

3.1Available

SAM 3.1

Segment Anything Model 3.1 from Meta. Unified foundation model for promptable segmentation in images and videos with 7x faster multi-object tracking.

Segmentationinteractivezero-shotmulti-objectopen-vocabularyOther

About this model

SAM 3.1 (Segment Anything Model 3.1) is Meta's latest foundation model for promptable image and video segmentation. Building on SAM 3, it introduces Object Multiplex — a shared-memory approach for joint multi-object tracking that achieves ~7x faster inference at 128 objects on a single H100 GPU with no accuracy loss. SAM 3.1 supports text prompts, visual prompts (points, boxes, masks), and open-vocabulary segmentation covering 50x more unique concepts than existing benchmarks. Improved on 6 out of 7 VOS benchmarks.

Plan the integration path

Avala batch inference currently supports SAM and YOLO. For other catalog models, use the provider workflow and connect compatible prediction outputs where supported.

Review inference support →

Capabilities

Task Types

Segmentation

Modalities

2D ImageVideo

Annotation Types Produced

PolygonSemantic Segmentation