Meta releases SAM 3.1, adding 7x faster multi-object tracking to vision foundation model
Meta has released SAM 3.1, an update to its Segment Anything Model that adds Object Multiplex, a shared-memory approach for joint multi-object tracking. The new version achieves approximately 7x faster inference when tracking 128 objects on a single H100 GPU while improving video object segmentation (VOS) performance on 6 out of 7 benchmarks.
Meta Releases SAM 3.1 with 7x Faster Multi-Object Tracking
Meta has released SAM 3.1, an updated version of its Segment Anything Model that introduces Object Multiplex, a shared-memory architecture designed for efficient joint multi-object tracking in videos.
Key Improvements
SAM 3.1 builds on SAM 3, Meta's unified foundation model for promptable segmentation that handles both images and videos. The new version maintains backward compatibility with SAM 3's core capabilities—detecting, segmenting, and tracking objects via text or visual prompts (points, boxes, masks)—while adding significant performance gains for tracking scenarios.
The Object Multiplex approach delivers approximately 7x faster inference when tracking 128 objects on a single H100 GPU, according to Meta's claims. The company states this speedup occurs without sacrificing accuracy. Additionally, SAM 3.1 shows improved video object segmentation (VOS) performance across 6 out of 7 benchmarks tested.
Concept Vocabulary Scale
SAM 3's underlying capability to handle open-vocabulary concepts remains central to SAM 3.1. The model can exhaustively segment all instances of concepts specified via short text phrases, handling over 50x more unique concepts than existing benchmarks, according to Meta.
Deployment Details
The SAM 3.1 model checkpoints are available on Hugging Face at facebook/sam3.1. Meta notes that there is no Hugging Face Transformers integration for this release. Users must access the SAM 3 GitHub repository for installation instructions, code examples, and full documentation. The model requires users to share contact information via Meta's privacy framework before access.
As of the latest data, the model has recorded 1,865 downloads in the previous month on Hugging Face. Currently, no inference providers have deployed SAM 3.1, meaning users must run the model locally or on their own infrastructure.
What This Means
SAM 3.1 positions Meta's segmentation foundation model for production use cases in video understanding and multi-object tracking—domains where inference speed directly impacts real-world feasibility. The 7x speedup at scale (128 objects) suggests the model targets enterprise applications in video analysis, autonomous systems, and robotics rather than single-object or lightweight use cases. The lack of inference provider deployment currently limits accessibility to developers with GPU infrastructure, though the open availability of checkpoints aligns with Meta's broader strategy of releasing foundational models for research and production adoption.
Related Articles
Runway releases Praxis-1 robot model with open weights under Apache 2.0, tops World Action Models on RoboLab-120
Runway has released Praxis-1, a robot control model, with open weights and source code under the Apache 2.0 license. It scores 46.2 percent on Nvidia's RoboLab-120 leaderboard, ahead of FLUX 3 Action at 42.9 percent, in tests Runway ran itself.
Meituan releases LongCat-2.5: 1.6T-parameter multimodal MoE with 48B active per token
Meituan's LongCat team has published LongCat-2.5, a native multimodal mixture-of-experts model with 1.6 trillion total parameters and approximately 48 billion activated per token. It targets long-horizon agentic tasks such as codebase development and full-project builds, and ships with W4A4 expert quantization for domestic AI accelerators.
Tencent Hunyuan releases Hy Realtime voice model with tool calling and singing, live in Yuanbao
Tencent Hunyuan announced Hy Realtime on November 13, a real-time voice model that decides on its own when to run a web search and can sing songs on request. It is live in Tencent Yuanbao voice calls and Tencent's customer service, with a Tencent Cloud API to follow. API pricing has not yet been disclosed.
Translated releases Lara Mini, a 7.2B-parameter MoE research model with 600M active parameters
Translated announced Lara Mini on November 12, 2026, a research model trained from scratch on 6 trillion tokens across English and 44 other languages, including all 24 official EU languages. The mixture-of-experts system has 7.2 billion parameters and activates 600 million per token.
Comments
Loading...