Pricing that scales with you
Choose VMs, serverless or on-demand GPUs, pay-as-you-go sandboxes, or BYOC.
Performance tier. Sandbox and VM prices are for 1 vCPU and 1 GiB of RAM.
Choose per workload for CPU functions, sandboxes and VMs.
Virtual Machines
Persistent Linux VMs for agents. Billed per second while running, with no compute charges while stopped.
Serverless
Compute is billed by the millisecond while your code is running.
An endpoint on an RTX 4090 with 2 physical cores (4 vCPU) and 16 GiB RAM costs $1.77/hr ($0.69 GPU + $0.76 CPU + $0.32 RAM, at the GPU-attached rates above). Billed by the millisecond, only while your code is running.
Sandboxes
Isolated, secure environments for running untrusted code at scale.
Pay by the millisecond, only when the sandbox is running.
A 2 vCPU / 8 GiB sandbox on Performance costs $0.18/hr, or $0.00005 per second.
On-demand
One flat price per machine. Includes vCPU, RAM, and NVMe drives.
One price per machine · Bare-metal or VMs · Need infiniband? See Clusters
8 hours on an H100 machine is $14.64. That price already includes the 26 vCPU, 200 GB RAM, and 1 TB NVMe drive.
Clusters
Multi-node GPU clusters with InfiniBand, reserved monthly or yearly. Every cluster connects to Beam, so one SDK runs your cluster, serverless workloads, and your own cloud.
Reserved · Multi-node · InfiniBand
Storage
Storage is not billed yet. Planned pricing is based on data stored, with no free allowance.
Planned billing covers retained VM disk, volume, and snapshot data, including while compute is stopped, and uses the same usage credits as compute. Empty disk capacity, shared base images, and internal copies are excluded. 1 GB = 1 billion bytes.
Bring your own cloud
Run workloads in your own cloud, including AWS, GCP, and Azure. Beam manages compute orchestration and the control plane. Pay for the compute using your cloud credits.
A g5.2xlarge instance on AWS (8 vCPU · 32 GB RAM) costs $0.44/hr in Beam fees. The instance itself is billed directly from your AWS account using your credits.
Ask an AI about pricing
Every rate on this page is also in pricing.md, so an AI assistant can estimate your costs.
Cost calculator
Pick a GPU (or not), add CPU cores and RAM, and set your monthly usage.
Plans
| Developer | Team | Growth | |
|---|---|---|---|
| Usage | |||
| Apps | Unlimited apps | Unlimited apps | Unlimited apps |
| Cloud Storage Volumes | Unlimited storage volumes | Unlimited storage volumes | Unlimited storage volumes |
| Custom Images | Unlimited custom images | Unlimited custom images | Unlimited custom images |
| Concurrent vCPUs | 30 vCPUs | 1,000 vCPUs | Custom |
| Concurrent GPUs | 5 GPUs | 50 GPUs | Custom |
| API Requests | Unlimited requests | Unlimited requests | Unlimited requests |
| Log Retention | 30 Days | 30 Days | 1 Year |
| Seats | 1 seat | 3 seats included, $25 per additional seat | Unlimited seats |
| Log Export | Included | Included | Included |
| Bandwidth | Included | Included | Included |
| Features | |||
| Cloud Storage Volumes | |||
| Secrets Manager | |||
| Deployment Logs | |||
| Export Your Code | |||
| Keep Warm | |||
| Autoscaling | |||
| Compliance | |||
| SOC2 Compliance | |||
| HIPAA Compliance | — | — | |
| Support | |||
| Community Support | |||
| Live Chat Support | |||
| Private Slack Channel | — | ||
Concurrent vCPUs is the total CPU running at once across VMs, sandboxes, and serverless apps, not a count of containers. A default VM uses 4 vCPUs, so the Developer plan can run 7 of them at once. Paused and stopped VMs don’t count.
Purchased credits don’t expire.
Running more than $10K/mo? Volume discounts are available. Book a call.
Beam is powering hands-down the best developer experience to run models on GPUs easily at scale. Best decision on the infra side for us this year so far.
Louis Morgner, Co-founder, AI lead @ Jamie
Frase is running language models exclusively on Beam and it was surprisingly easy to migrate, less maintenance, and is saving us money.
Frankie L., CTO at Frase
Powering products and agents at Coke, Magellan, StratumAI, Raven, Ogilvy, Invoke, Frase, Jamie and EdgeImpulse.
Frequently asked questions
How much does storage cost?
Storage is not billed yet. Planned: $0.0001 per GB-hour of data stored, with no free allowance. Planned billing covers retained VM disk, volume, and snapshot data, including while compute is stopped, and uses the same usage credits as compute. Empty disk capacity, shared base images, and internal copies are excluded. 1 GB = 1 billion bytes.
I don't want serverless. Can I use Beam while keeping the server running 24/7?
Yes. By default, apps will spin down automatically after each request, but you can control how long your apps are active. Many customers with latency-sensitive workloads choose to keep their servers running 24/7 on Beam.
Am I billed for cold start?
No. We only charge for the time to load your application code. We don’t charge for the time to spin up a server or load your container image.
I have an existing Docker image. Can I use it on Beam?
Yes. You can import any base image from a third-party image registry onto Beam.
How secure is Beam?
Every workload is isolated from the others. VMs run as microVMs with their own kernel. Sandboxes and serverless apps run in isolated containers, and a sandbox can run in a microVM with use_vm=True. We also offer a self-hosted product that runs entirely in your own environment, so no data leaves your VPC.
Can I run Beam in my own cloud account?
Yes. With Bring Your Own Cloud, Beam manages workloads directly on instances in your AWS or GCP account, so you can use your existing cloud credits and negotiated rates. You pay Beam a flat management fee of $0.019/hr per vCPU and $0.009/hr per GB of RAM — the compute itself is billed to your cloud account.
Do you charge for egress or bandwidth?
No. There are no egress or bandwidth fees on any plan.
Give Your Agent Superpowers
Start shipping today and bring your vision to life