SAM 3.1 vs Ternary Bonsai

SAM 3.1 is Meta's latest update to the Segment Anything Model family, released March 27 2026 as a drop-in replacement for SAM 3. The core innovation is object multiplexing: where the previous model required a separate processing pass for each tracked object, SAM 3.1 processes all tracked objects together in a single shared-memory pass, eliminating redundant computation across the decoder. The result is a doubling of throughput for videos with a medium number of objects—from 16 to 32 frames per second on a single H100 GPU—without sacrificing tracking accuracy. For applications like sports analytics, surveillance, or video editing that track 5–20 objects simultaneously, this makes real-time deployment on commodity cloud hardware feasible for the first time. SAM 3.1 inherits SAM 3's open-vocabulary segmentation capability (segmenting objects described by text prompts), which achieved 75–80% of human performance on the SA-CO benchmark covering 270K unique concepts. The model checkpoint is available on Hugging Face at `facebook/sam3.1`, and the codebase supports fine-tuning via the facebookresearch/sam3 repository. Meta released SAM 3.1 under a research license with commercial use provisions similar to its predecessors.

PrismML's Ternary Bonsai is a family of ultra-compressed language models using 1.58-bit weights — meaning every parameter is stored as -1, 0, or +1, with no higher-precision layers anywhere in the architecture. The line-up covers 8B, 4B, and 1.7B parameter models. The flagship 8B model fits in 1.75 GB of RAM, a 9x reduction versus a 16-bit baseline. Unlike earlier 1-bit experiments that felt like a party trick with serious capability regressions, Ternary Bonsai 8B outperforms PrismML's own prior 1-bit Bonsai 8B by 5 points on average across standard benchmarks. The team also ships WebGPU inference, so the 1.7B model runs entirely in a browser tab. This is the first time a production-quality chat model has run with no server at all. The real-world use case is edge and offline deployment: medical devices, air-gapped government systems, consumer apps that need to work without a signal. At 1.75 GB, the 8B model fits on the GPU RAM of a six-year-old gaming laptop. PrismML is positioning this as the foundation for truly offline AI — a credible claim if the capability benchmarks hold up under real-world testing.