Dedicated AMD and NVIDIA GPUs, launched from the console or the CLI in minutes. Pay per second while they run.
ROCm images run on AMD GPUs and CUDA images on NVIDIA GPUs. The console lists the models you can launch now.
Metered from the moment the container starts. Waiting for capacity and pulling the image are free, and a stopped instance is not billed for compute.
The console shows what is available and holds it when you launch. The CLI launches the same instance from a script.
Every instance runs on dedicated GPUs, and each GPU comes with its own share of CPU and memory. Launch 1, 2, 4 or 8 on a single node, so a job that outgrows one card scales without a second machine to wire up.
Pick the GPU that fits the stack you already have: ROCm images run on AMD GPUs and CUDA images on NVIDIA GPUs. Start from a maintained template or bring your own image.
A fine-tune runs for hours or days and then stops. Billing is per second and ends when the container does, so between jobs you pay for stored volumes and nothing else.
Adapt an open model to your own data. Start from a PyTorch template or your own image, write checkpoints to a network volume, and stop the instance when the run ends.
Take up to eight GPUs in one instance for a long run. Checkpoints on a network volume survive a stop, a restart and a deleted instance.
Serve a model with vLLM or SGLang behind a public HTTPS port and test it with your own traffic. When the workload settles into production, move it to managed GPU for an SLA.
Submit one command as a Run Job from the CLI. It streams output back and releases the GPU when the command exits. You pay only while the command runs.
Waiting for capacity and pulling the image are free. A stopped instance is not billed for compute. The rate you launch at holds for that instance, and there is no contract.
Add credit by card and pay per second while the container runs. GPU rates depend on the model and the count, and the console shows the rate before you launch.
Rates as of 19 November 2026. Run featherless-cloud pricing list to print the current rate for every GPU model.
Use your Featherless account or create one, then add credit by card. One hour of an instance's rate is enough to start it.
PyTorch, vLLM and SGLang templates ship with SSH and JupyterLab, or bring your own linux/amd64 image. Choose the GPU model, take 1, 2, 4 or 8, and mount a network volume for anything you want to keep.
SSH in as root, open the browser terminal, or work in JupyterLab. Stop the instance when the job ends and compute billing ends with it.
Match the GPU to your libraries. CUDA images need NVIDIA GPUs and ROCm images need AMD GPUs. On AMD GPUs, PyTorch's torch.cuda API drives the GPU through ROCm, so code without CUDA-only dependencies generally runs as written. The maintained PyTorch, vLLM and SGLang templates are built on ROCm. If you are unsure about a library, ask in Discord before you add credit.
The console lists every model you can launch now, with its memory and its rate, and holds the capacity when you launch. The CLI prints the same list with featherless-cloud pricing list. Billing is per second, so ten minutes of your own job costs a sixth of the hourly rate.
No. One instance or Run Job uses up to eight GPUs on a single node. You can run several side by side, up to 16 GPUs per organization by default, and ask in Discord for a higher limit.
Not on GPU Cloud. That is fine for a training run and wrong for production traffic with a latency target. If you need an uptime commitment, managed GPU is the product.
We maintain PyTorch, vLLM and SGLang templates, each with SSH and JupyterLab. Everything after launch is yours: the model, the settings and the job. If you want a model catalog behind an API, that is our serverless inference product.
Not from a running instance. It sits on dedicated compute until you stop it or your credit runs out. If you stop it, starting again needs a free GPU of the same model, which can fail when capacity is short. Run Jobs are different: they stop at 24 hours and can be interrupted to make room for other work, so allow retries and write checkpoints to a network volume.
No. It is your data and we do not read it. We monitor metadata only, to catch abuse such as someone attacking a third party from our network. Files on a network volume stay on our disks until you delete the volume, so we have legal obligations and would cooperate with a lawful request.
Usage is metered per second from the moment the container starts. Waiting for capacity and downloading the image are not billed. A stopped instance is not billed for compute, and its container disk bills only while it runs. Network volumes bill for their full size from creation until you delete them.
The container disk is erased on every stop and restart, installed packages included. Keep datasets, checkpoints and the Hugging Face cache on a network volume. A volume keeps its data when the instance stops or is deleted, can grow while mounted, and can be shared by several instances.
There is no commitment. An instance starts once your credit covers one hour of its rate. After that you pay per second.
The region and egress pricing. If either decides it for you, ask in Discord or email gpu@featherless.ai before you add credit.
Sign in with your Featherless account, add credit and launch. The quickstart covers sign-in, launch, running code and clean-up.
If you are moving an inference workload or need more than 16 GPUs, talk to us about managed GPU or a higher limit.