According to AI at Meta on X, Meta has released SAM 3.1, a drop-in update to SAM 3 that adds object multiplexing to significantly improve video processing efficiency without sacrificing segmentation accuracy. As reported by AI at Meta, the update is intended to enable high‑performance video understanding on smaller GPUs, opening opportunities for cost-effective, real-time applications in video editing, robotics perception, AR capture, and retail analytics. According to AI at Meta, object multiplexing allows multiple object tracks to be processed concurrently within shared compute, reducing per-object latency and GPU memory footprint while maintaining the quality levels established by SAM 3. As reported by AI at Meta, Meta is sharing the update with the community, positioning SAM 3.1 as a practical upgrade path for developers seeking scalable video instance segmentation and tracking on constrained hardware.

According to AI at Meta, the core innovation in SAM 3.1 is object multiplexing, enabling the model to track up to 16 objects in a single forward pass, whereas earlier versions required a separate pass per object, eliminating redundant computation and reducing inference latency and cost. As reported by AI at Meta, batching objects in one pass improves throughput for multi-object video segmentation and tracking, a critical workflow for retail analytics, robotics perception, sports broadcasting, and video editing. According to AI at Meta, this architectural change consolidates feature extraction, which can cut per-frame GPU calls and memory transfers, creating opportunities to scale real-time multi-object tracking with fewer accelerators.