Making It Easier & Cheaper to Deploy with Roboflow

Roboflow announced on September 18, 2026 that its Serverless API moves to per-image pricing, its Core Plan gains a free tier starting at $0 with 10 credits, and self-hosted inference via Roboflow Inference becomes free on any plan and workspace. The company said state-of-the-art object detection models such as RF-DETR Nano, Small, and Medium will be more than 40% cheaper to run than before, and that 9 in 10 workspaces running models on Serverless will be cheaper than self-hosting on another cloud. Core Plan also now includes private projects, unlimited seats, and model weight download, with volume discounts starting at 250,000 images per month.

Our mission is to make the world programmable, and accelerate vision AI into the world. That means building an ecosystem where developers can experiment, founders can turn prototypes into businesses, and people who know a problem firsthand can use vision AI to solve it. We're announcing four major changes to deliver on that mission: 1. Serverless API moves to a cheaper, per-image pricing model 2. The Core Plan now starts at $0 and includes 10 credits 3. Self-hosted inference using Roboflow Inference becomes free 4. All Core Plans include private projects, unlimited seats, and model weight download Serverless API Pricing Gets Cheaper & Predictable Starting today, September 18, 2026, the Serverless API bills at a per-image rate depending on the model you use. Per-image pricing means you know the cost before you run any model: price per 1,000 images, times your volume. Our prior pricing model for Roboflow Cloud's Serverless API https://docs.roboflow.com/deployment/roboflow-cloud/serverless-api?ref=blog.roboflow.com was based on GPU hours on execution, which meant the same job cost a different amount depending on cold starts, model size, queuing, inputs, etc. This led to the cost of deploying a vision solution into production being highly variable, and with this new change, we are delivering on what users requested to provide easy to forecast pricing. Additionally, Roboflow Serverless API is becoming even more cost-efficient. With our experience processing more than 100 billion images and video frames just in the past year, and our relentless engineering efforts https://blog.roboflow.com/serverless-inference-a-thousand-models-on-a-shared-gpu-fleet/ in making Serverless incredibly efficient to run, we get to pass these savings onto our customers. In our benchmarks, 9 in 10 workspaces running models on Serverless will be cheaper than self-hosting on another cloud . For some models, like state-of-the-art object detection models like RF-DETR Nano, Small, and Medium, it will become more than 40% cheaper to run than before. Roboflow Serverless includes popular models like RF-DETR, YOLO, SAM 3, QwenVL, and more constantly being added. It becomes even cheaper to run at scale with volume discounts starting at 250,000 images per month. See our per-image model pricing in our docs https://docs.roboflow.com/deployment/roboflow-cloud/serverless-api/model-pricing?ref=blog.roboflow.com , and get started https://app.roboflow.com/?ref=blog.roboflow.com . Core Plan Gets a Free Tier Our Core Plan, previously a paid plan, will now get a Free Tier, which comes with all the features and tools to get a vision use case into production. This includes private projects and models, as well as neural architecture search NAS , model evaluation, training analytics, preprocessing, and model weight download. The Free Tier will come with 10 credits, which is enough to train ~30 models or run 80,000 inferences with the state-of-the-art RF-DETR Nano model entirely for free, and the ability to buy more credits as you need them. Additionally, Core Plan now includes unlimited seats. As the technical barrier to using Roboflow lowers with Roboflow Agent, our MCP, and other advancements, we want to unlock the capability for as many people as possible to collaborate, iterate, and build with Roboflow. Self-Hosted Inference Is Free On any plan and on any workspace, using Roboflow Inference to run models or Workflows locally will no longer consume credits. We’ve long open-sourced our package for running vision models at scale, Roboflow Inference https://inference.roboflow.com/?ref=blog.roboflow.com . It’s the best way to run cutting-edge models while processing video streams, optimizing compute resources, and managing dependencies out of the box. In the past, it consumed credits on a workspace. No longer. Opening up powerful, efficient Inference to the masses will accelerate our mission to democratize computer vision. For companies needing expanded capabilities at the edge, our Advanced Edge Deployments product which includes features to support running in Enterprise OT networks, advanced HMIs, compatibility with industrial cameras and protocols, and functionality for managing fleets of edge devices , and commercial model weights licensing https://roboflow.com/licensing?ref=blog.roboflow.com for models that need it are available on our Enterprise tier. For Existing Customers Your existing plan is getting more features at no added cost. Monthly customers will continue to pay the same price. Annual customers keep the term they paid for. If you pay for seats today, those charges will stop, and no users will be removed. Public Plan workspaces will move to the Core Plan Free Tier. Users can continue to create open source projects to contribute to Roboflow Universe and have access to Model Evaluation, training analytics, advanced preprocessing, and creating new private projects. If you’re already using Roboflow, nothing changes until we email you. You’ll receive a clear email breaking down your specific plan’s explanation of your credits, your pricing, and what’s new. The World Is Ready For What You’ll Build The next important vision AI application could come from a two-person startup, a developer experimenting after work, or an engineer who knows exactly what their factory needs. We want each of them to be able to start building. That is what being an ecosystem builder means to us: making it easier for other people to create. More experiments become useful applications. More applications become products and businesses. More of the world benefits from what vision AI can do. Our business grows as teams choose Roboflow to run at scale. Making it easier to get started aligns our success with yours. You bring the idea and the knowledge of the problem; we’re building the tools and infrastructure to help you take it into the world. Vision AI is already doing useful work on plant floors, in rail yards, and in fields. We want to help the next generation of engineers bring that kind of ambition to problems everywhere. The new free Core Plan https://roboflow.com/pricing?ref=blog.roboflow.com is an invitation to build. Start with a camera, a few hundred images or a video stream, and a problem worth solving. Bring your team. Build your first model. Take it into the world. The next great vision AI company could start with your first project. Let’s build it Cite this Post Use the following entry to cite this post in your research: Leo Ueno https://blog.roboflow.com/author/leo/ . Sep 18, 2026 . Making It Easier & Cheaper to Deploy with Roboflow. Roboflow Blog: https://blog.roboflow.com/making-it-easier-cheaper-to-deploy-with-roboflow/