| Main focus | Computer vision; labeling, training, and deployment for images and video | Physical AI data; curation and annotation for robotics, autonomous vehicles, drones, and industrial teams | Frontier AI labs; RL environments, evals, and expert data for post-training, plus enterprise agents |
| Data and task types | Computer vision; object detection, instance and semantic segmentation, classification, keypoints, OCR, depth estimation, and multimodal tasks | Broad; image, video, 3D and LiDAR, audio, documents and text, and DICOM / NIfTI medical imaging | Broad multimodal; image, video, text, documents, geospatial, audio, conversational text, and HTML, plus LLM evaluation editors |
| AI-assisted labeling | Yes, Auto Label labels the classes you describe in a text prompt; Smart Polygon powered by Meta AI's Segment Anything; Label Assist pre-annotates with your own trained models | Yes, SAM 3 segmentation, SAM 2 tracking across video, interpolation, and object tracking | Yes; Foundry runs foundation models to generate labels, AutoSegment uses Meta's Segment Anything, and pre-labels from your own models can be imported for review |
| Review and QA workflow | Yes, Admin, Labeler, and Reviewer roles with assignment tracking; annotation history with version revert | Yes, customizable workflows with annotate, review, consensus, and agent stages | Yes, custom labeling and review workflows and a project performance dashboard on Starter and above |
| Managed labeling | Available; a managed labeling service is available | Yes, Data-as-a-Service for expert annotation and data collection | Yes; the Alignerr expert network, plus managed labeling with a quality SLA on Enterprise |
| Data curation | Built in; dataset versioning, preprocessing, and augmentation | Yes, embedding-based search and model-in-the-loop curation to find edge cases | Catalog; connects to 25+ cloud sources, with vector and traditional search and curation |
| Export formats | 40+ formats, including COCO, YOLO, and Pascal VOC | Encord JSON and COCO; YOLO export when enabled on the account | JSON; per-data-type JSON exports; COCO, YOLO, and Pascal VOC exports not documented |
| Train your own vision model | Yes, RF-DETR, YOLO (v8 / v11 / v26), Detectron2, and more | Limited; in-app training for detection, segmentation, and classification, with architectures not named in docs | Bring your own; custom models plug in for pre-labeling and error analysis; native training not documented |
| Export and own model weights | Yes | PyTorch only; weights downloadable per training iteration | Not documented; no native training to export from |
| Edge deployment | Yes, NVIDIA Jetson, GPU servers, Raspberry Pi, x86, or Roboflow AI1, online or fully offline | Not documented | Not documented |
| Platform deployment options | Managed cloud, customer VPC, on-premises, and fully air-gapped / offline | Cloud, VPC, on-prem; US or EU hosting, with private VPC and on-premises on Enterprise | Labelbox-hosted on Google Cloud, with customer-hosted assets via signed URLs or delegated access; on-premises and air-gapped not documented |
| Self-hosted inference server | Yes, open-source inference server for custom and foundation models | No open-source, self-hostable inference server is documented | Not documented |
| Compliance | SOC 2 Type II, HIPAA with BAAs, PCI DSS | SOC 2, HIPAA, GDPR | SOC 2 Type II, HIPAA compliance program, GDPR, and CCPA; PCI DSS and NIST 800-171 also listed on labelbox.com |
| Pricing | Published plans; every plan, including Free, includes self-hosted inference, model weights download, and a private workspace | Starter, Team, and Enterprise plans; prices not published | Free plan (up to 30 users and 50 projects), Starter, and custom-priced Enterprise; labeling services priced per hour |