Roboflow
End-to-end computer vision platform: AI labeling, datasets, training (RF-DETR, YOLO, SAM 3), Workflows with VLMs and self-hosted Inference
Overview
Roboflow is an end-to-end computer vision platform: you upload images, label them, manage datasets, train a model, and deploy it to the cloud or your own hardware. Its AI features include AI Labeling (Label Assist, Smart Polygon, Box Prompting and Auto Label with foundation models), Workflows that chain models such as RF-DETR, YOLO, SAM 3 and the Qwen3-VL vision-language model with logic and integrations, and a Roboflow Agent that helps take an idea to a deployed vision app. It also maintains open-source projects: the Supervision toolkit, the RF-DETR detection model (Apache-2.0) and the Inference server, which runs models on your own hardware free on every plan. Brad Dwyer and Joseph Nelson founded it in 2020 (Y Combinator S20). Roboflow says more than one million developers use it, including engineers at more than half of the Fortune 100, and it raised a $40M Series B led by GV in November 2024.
The Verdict
Who Should Use Roboflow?
Best For
- Teams that need labeling, training and deployment in one place
- Fast prototypes of detection or segmentation apps
- Edge deployments that run models locally with Inference
Not Ideal For
- Teams on a tight budget with heavy cloud training needs
- Projects that want a fully open-source stack only
What's Great
- Ease of use is the top theme in G2 reviews
- Efficiency is the next most cited pro
- Support quality praised by reviewers
Watch Out For
- "Expensive" is the most common G2 complaint
- Forum users report surprise credit charges on training and Workflows
- "Lack of features" is the next most common complaint
Pricing
View all features & details
Key Features
- AI Labeling: Label Assist, Smart Polygon, Box Prompting, Auto Label
- Hosted training, including RF-DETR and YOLO models
- Workflows combining models, VLMs, logic and integrations
- Roboflow Agent for building vision apps
- Serverless API, dedicated deployments, cloud streaming
- Self-hosted Inference server, free on every plan
How Credits Work
- Credits pay for cloud inference, labeling, training and Workflow steps
- RF-DETR Nano: 8,000 images per credit; SAM 3: 2,000 images per credit
- Prepaid credits from $4, on-demand $6; included credits do not roll over
Real-World Usage
Adoption
- 1M+ developers, engineers at more than half of the Fortune 100
- Series B backers include GV, Craft Ventures and Y Combinator
Open Source
- Supervision toolkit (MIT), widely starred on GitHub
- RF-DETR (Apache-2.0) and Inference, all pushed daily
How It Compares
| Feature | Roboflow | Ultralytics |
|---|---|---|
| Main job | Labeling, datasets, training and deployment | Open-source YOLO models plus a platform |
| License | Commercial SaaS; open-source libraries | AGPL-3.0 or Enterprise |
| Self-host inference | Yes, Inference is free | Yes |
| Entry paid plan | $39/mo (Core) | $29/seat/mo |
Sources: Roboflow · Pricing · About · Series B (Nov 2024) · G2 reviews · see also Ultralytics and LandingAI
Is this your tool? Add a badge