Meta has listed Segment Anything Model (SAM) 3.1 on its developer platform, Meta Model API, describing it as the company’s top perception model for detecting, segmenting and tracking objects in images and video. Meta says the model works zero-shot: a developer types a short phrase naming an object, and SAM 3.1 returns a box and a pixel-precise mask, then tracks that object through video while keeping its identity, without any fine-tuning or training data.
That “leading” label and the claim of “production throughput on day one” are Meta’s own; the developer page includes no independent benchmark results.
SAM 3.1 shares billing, docs and API keys with Meta’s other models on the same platform, including its reasoning model Muse Spark and its transcription model Muse Voice Transcribe, so it can plug into an existing pipeline. Images cost $2.50 per thousand processed. Video tracking runs $0.20 per thousand frames. Developers on OpenAI SDK compatible clients can point them at Meta Model API and get a first request running in under five minutes, Meta says.
For teams doing manual video annotation, SAM 3.1 is worth a pricing comparison against existing segmentation vendors before the next contract renewal.
According to Meta’s developer documentation for Segment Anything Model 3.1.
