AI code-execution sandbox

Beam

Beam is a serverless sandbox and GPU platform with sub-second cold starts, per-second billing and no subscription floor, notable for offering GPUs inside the sandbox (H100 at $1.74/hr).

Last verified 2026-08-09

Beam runs sandboxes and serverless GPU workloads through one SDK, billed per second with a $30/month free-credit allowance and no subscription floor. Its distinguishing feature in this category is affordable GPU inside the sandbox: an H100 is $1.74/hr, well below the roughly $3.95/hr common elsewhere. It can also run in your own cloud (BYOC) for a flat management fee.

Key specs

IsolationContainer
Pricing modelPer-second usage (no subscription floor)
GPUH100 $1.74/hr, A100 80GB $1.30/hr, RTX 4090 $0.42/hr
Session limitConfigurable
PersistenceEphemeral sandboxes; volume storage
LanguagesPython, any (container)
MCP supportn/a
DeploymentManaged cloud + BYOC (self-host)
Free tier$30/mo free credits, no subscription floor
Sandbox (2 vCPU / 8 GiB)$0.319/hr
H100 GPU$1.74/hr
BYOC$0.019/vCPU-hr fee

Strengths

  • GPU in the sandbox at low rates (H100 $1.74/hr)
  • Per-second billing, no subscription floor
  • $30/mo free credits
  • Bring-your-own-cloud option

Trade-offs

  • CPU-only rate is not the cheapest at steady high volume
  • Container isolation is lighter than a microVM

Best for: Agents that need GPUs inside the sandbox, or spiky usage a subscription floor would penalize.

How Beam works

Beam runs sandboxes and serverless GPU workloads through a single SDK, billed by the second and only while your code runs. What separates it in this category is GPU inside the sandbox: the same API that runs CPU code can attach an H100, A100 or consumer cards, so accelerated inference sits next to the agent's code instead of on a separate service. There is no subscription floor, and $30 in monthly credits covers experimentation. Beam can also run in your own cloud account (BYOC), where you pay Beam a flat orchestration fee and the compute bill goes to your provider.

Pricing in practice & watch-outs

The headline is GPU value: an H100 is $1.74/hr, against roughly $3.95/hr on several managed sandboxes, and a 2 vCPU / 8 GiB sandbox is about $0.319/hr. For pure CPU work at steady, high volume a per-vCPU-priced competitor can come out cheaper, so Beam is strongest when you either need a GPU, have spiky usage that a subscription floor would tax, or want to self-host on cloud credits you already hold. Isolation is container-based, lighter than a microVM boundary, which is worth weighing for fully untrusted code.

Head-to-head

Alternatives to Beam

See the full Beam alternatives comparison →

Sources: Beam pricing ↗ · Beam ↗

Is Beam right for your workload?

Tell us your use case and we'll confirm the fit, flag anything to watch for, and get you real pricing from the Beam team. No spam.