Beam
Serverless GPUs and sandboxes for AI workloads
Last reviewed Mar 14, 2026
Beam is an open-source serverless cloud platform for AI and ML workloads, offering serverless GPUs, secure sandbox environments, flat-priced on-demand machines, and reserved multi-node clusters. Workloads can also run in the customer's own AWS, GCP, or Azure account with Beam handling orchestration.
Available GPUs
Hourly on-demand pricing. Click column headers to sort.
Prices last updated: August 21, 2026
Beam pricing by GPU
Configurations, price rank, and alternatives for one GPU at a time.
Pros & Cons
Advantages
- Pay-per-millisecond billing with no charges for cold starts
- Ultrafast boot times with sub-second cold starts via memory snapshots
- GPU selection spanning RTX 4090 through B200, plus InfiniBand cluster nodes
- On-demand machines priced flat per machine, including vCPU, RAM, and NVMe storage
- Secure sandbox environments ideal for AI agents and code execution
- Open-source core engine (beta9) available for self-hosting
- Workloads can run in your own AWS, GCP, or Azure account, avoiding lock-in
- Unlimited apps, cloud storage volumes, and custom images on every plan
- Developer-friendly Python and TypeScript SDKs with easy deployment
Limitations
- Serverless GPUs other than the RTX 4090 require a committed-spend agreement
- GPU container concurrency is capped on the lower plan tiers
- Multi-node InfiniBand clusters are sales-only with no published pricing
- Log retention limited to 30 days outside the top tier
- Newer platform with smaller community
Key Features
Pay-per-millisecond Billing
Only charged when your code runs, no charges for cold starts or server spin-up
Sub-second Cold Starts
Memory snapshots and GPU checkpoint restore bring containers back in seconds, which Beam reports as up to 35x faster than a traditional cold boot
Secure Sandbox Environments
Run untrusted code safely in isolated environments for AI agents and code interpreters
Stateful, Persistent Runtimes
Pause and resume sessions while maintaining filesystem, memory, and running processes
Elastic GPU Scaling
Scale to zero when idle, burst to thousands of containers in seconds
Custom Docker Images
Bring your own Docker images for full environment control, including running the Docker daemon inside containers
Bring Your Own Cloud
Connect AWS, GCP, Azure, or bare-metal accounts and run workloads across them from the same SDK
Open Source
Core engine (beta9) is open-source and can be self-hosted
Compliance Certifications
SOC 2 and HIPAA compliance available on higher plan tiers
Compute Services
Inference Endpoints
Deploy high-performance inference endpoints with custom models
- Custom model hosting on GPU or CPU
- Pre-loading models for fast response times
- Realtime and streaming capabilities
Sandboxes
Secure code execution environments for AI agents
- Run untrusted LLM-generated code safely
- Snapshot and restore complete system state
- File system operations and process management
Task Queues
Run large-scale workloads with distributed processing
- Audio processing pipelines
- Data processing at scale
- Queue management and status tracking
Containers
Host arbitrary containers as cloud services
- Import Docker images from any registry
- Web service hosting
- Streamlit and Gradio UI support
On-Demand Machines
Bare-metal or virtual GPU machines billed at one flat price per machine
- vCPU, RAM, and NVMe drives included in the machine price
- Choice of bare-metal or VM delivery
- Nine GPU types from RTX 4090 through B200
Reserved GPU Clusters
Multi-node GPU clusters with InfiniBand interconnect, reserved monthly or yearly
- 8-GPU nodes with InfiniBand networking
- H100, H200, B200, and B300 SXM configurations
- Managed through the same SDK as serverless workloads
Bring Your Own Cloud
Beam orchestrates compute inside your own AWS, GCP, or Azure account
- Beam manages the control plane and compute orchestration
- Compute billed directly by your cloud provider using your credits
- Egress and bandwidth included in the management layer
Pricing Options
| Option | Details |
|---|---|
| Pay-per-millisecond | Serverless GPU, CPU, and sandbox usage billed by the millisecond with no minimum charges |
| Committed Spend | Serverless access to GPUs beyond the RTX 4090 is arranged through a committed-spend agreement |
| On-Demand Machines | Flat hourly price per machine that already includes the vCPU, RAM, and NVMe storage |
| Reserved Clusters | Multi-node InfiniBand clusters reserved monthly or yearly, quoted through sales |
| Bring Your Own Cloud | Per-vCPU and per-GB management fee on top of compute billed directly by your own cloud provider |
| Storage | Persistent volumes and snapshots included up to a capacity threshold, then billed per GB per month |
| Monthly Plans | Developer plan free with usage, Team plan with a monthly base fee plus usage, and a contact-sales Growth tier |
| Free Monthly Credits | All plans include monthly free credits that refresh each month |
| Volume Discounts | Discounts available for high monthly usage, arranged through sales |
Availability & Support
Regions
30+ regions spanning the US, EU, Asia-Pacific, and Canada, with workloads routed across clouds and regions in real time
Support
Documentation, public Slack community, and a status page; live chat support on paid plans and a private Slack channel on the top tier
Getting Started
- 1
Create an account
Sign up on the Beam platform to receive monthly free credits
- 2
Set up Python environment
Create a virtual environment and install the Beam SDK
- 3
Authenticate with API token
Configure your API token to connect to Beam
- 4
Deploy your first app
Run a function locally, then deploy it as a web endpoint with the Beam CLI
Compare Providers
Find the best prices for the same GPUs and models from other providers
