visionfamily
DINOv3
Meta self-supervised vision backbone family for general image features used for classification, detection, segmentation and depth. Task benchmarks not fetched.
Levelfamily
Classvision
Parameters6.72B
Contextnot in the notes
Released13 Aug 2025
Opennessopen weights restricted
Runs onhyperscale cluster, multi-GPU node, single datacentre GPU, CPU server, embedded GPU
Statusactive
Lineage
successorfinetune
DINOv2 is the parent (successor)high confidencemodel cardDINOv3 ConvNeXt-B belongs to this familyvision13 Aug 2025DINOv3 ConvNeXt-L belongs to this familyvision13 Aug 2025DINOv3 ConvNeXt-S belongs to this familyvision13 Aug 2025DINOv3 ConvNeXt-T belongs to this familyvision13 Aug 2025DINOv3 ViT-7B belongs to this familyvision13 Aug 20256.72BDINOv3 ViT-B belongs to this familyvision13 Aug 2025DINOv3 ViT-H+ belongs to this familyvision13 Aug 2025840MDINOv3 ViT-L belongs to this familyvision13 Aug 2025DINOv3 ViT-S belongs to this familyvision13 Aug 202521MDINOv3 ViT-S+ belongs to this familyvision13 Aug 2025DINOv3 SAT-493M ViT-7B belongs to this familygeospatialAug 20256.7BDINOv3 SAT-493M ViT-L (distilled) belongs to this familygeospatialAug 2025300MWhere this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 4 models. Confidence and evidence are the notes' own.
Identity
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.
Architecture and Size
- Architecture
- dense transformer Self-supervised ViT; ViT-7B teacher, distilled ViT-S/S+/B/L/H+ and ConvNeXt T/S/B/L
- Total parameters
- 6.72B sources
- Active parameters
- –
- Checkpoint parameters
- –
- Note
- ViT-S 21M to ViT-H+ 840M, ViT-7B 6,716M; ConvNeXt 29M to 198M. Training data LVD-1689M (1.7B images)
- Context length
- not in the notes
- Input
- image
- Output
- not in the notes
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).
Openness and Access
- Openness
- open weights restricted
- Licence
- DINOv3 License sources
- Access routes
- open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.
Deployment
- Runs on
- hyperscale cluster, multi-GPU node, single datacentre GPU, CPU server, embedded GPU
- Footprint
- ViT-7B: GPU cluster or large GPU; distilled S/B and ConvNeXt-T: edge/CPU feasible (judgement)
- Runtimes
- not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.
Indonesia
- Indonesian support
- not stated
- Regional languages
- not in the notes
- Jakarta hosting
- unknown
- Languages
- not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.
Benchmark Results
The notes record no benchmark result for this model.
Industry Use Cases
No industry use case links this model.
Links
Models 13
DINOv3 ViT-7BMeta, visionFamilyDINOv2Meta, visionsuccessorDINOv3 ConvNeXt-BMeta, visionFamilyDINOv3 ConvNeXt-LMeta, visionFamilyDINOv3 ConvNeXt-SMeta, visionFamilyDINOv3 ConvNeXt-TMeta, visionFamilyDINOv3 SAT-493M ViT-7BMeta, geospatialFamilyDINOv3 SAT-493M ViT-L (distilled)Meta, geospatialFamilyDINOv3 ViT-BMeta, visionFamilyDINOv3 ViT-H+Meta, visionFamilyDINOv3 ViT-LMeta, visionFamilyDINOv3 ViT-SMeta, visionFamilyOrganisations 1
Sources
- huggingface.co/facebook/dinov3-vit7b16-pretrain-sat493m https://huggingface.co/facebook/dinov3-vit7b16-pretrain-sat493m
- Meta Releases DINOv3: Advanced Self-Supervised Vision Transformer with 6.7B Parameters for Superior Image Embeddings | AI News Detail blockchain.news https://blockchain.news/ainews/meta-releases-dinov3-advanced-self-supervised-vision-transformer-with-6-7b-parameters-for-superior-image-embeddings
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "DINOv3", https://aurion.semai.id/models/meta-dinov3/, records of October 2026.