Segment Anything Models by AI at Meta
2 models
Last refreshed 2026-07-11. Next refresh: weekly.
About
Meta's Segment Anything foundation-model family for promptable image and video segmentation, including the distinct SAM 2 and SAM 3 releases.
Decision facts
- Best fit
- image segmentationvision and multimodal work
- Capability starting point
- Segment Anything Model 2 with multimodal inputs
- Lowest tracked input
- Not tracked
- Closest related family
- Chameleon
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
2 in view
Segment Anything Model 2Current
Use when the workload needs image segmentation and multimodal inputs.
Unknown releaseimage segmentationmultimodal inputs
Segment Anything Model 3Current
Use when the workload needs image segmentation and multimodal inputs.
Unknown releaseimage segmentationmultimodal inputs
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Segment Anything Model 2 | Use when the workload needs image segmentation and multimodal inputs. | Unknown release | image segmentationmultimodal inputs | Current |
| Segment Anything Model 3 | Use when the workload needs image segmentation and multimodal inputs. | Unknown release | image segmentationmultimodal inputs | Current |
Release Timeline
1 release groupUnknown release
2 current
Segment Anything Model 2
Currentimage segmentationmultimodal inputs
Segment Anything Model 3
Currentimage segmentationmultimodal inputs
Specifications(2 models)
| Model | Released | Vision | Multimodal |
|---|---|---|---|
| Segment Anything Model 2 | — | Yes | Yes |
| Segment Anything Model 3 | — | Yes | Yes |
