Serverless GPU and microVM sandboxes for AI agents, coding loops, and RL rollouts with $30/mo free Starter credits, elastic burst billing, and filesystem/memory snapshots.
Visit Modal SandboxesBest for GPU-heavy coding agents, RL rollouts, and teams using elastic burst limits to avoid idle-wait overbilling.
Serverless gVisor and VM sandboxes with massive GPU parallelism and elastic idle billing ($4.44 per 1,000 85%-idle runs), offset by a 3x non-preemptible rate multiplier and OAuth+card signup gate.
$0/mo Starter ($30/mo free credits, card req), $250/mo Team ($100/mo credits), 3x base rate on CPU Sandboxes ($0.07074/vCPU-hr)
Modal Sandboxes run untrusted AI-generated code, coding agents, and heavy GPU workloads inside isolated serverless containers (gvisor) and microVMs (vm, up to 512 GiB root with Docker-in-Sandbox).

Modal owns the heavy GPU and reinforcement-learning end of the sandbox market, combining fast parallel containers with H100/A100/B200 burst capacity, Filesystem Snapshots (30-day TTL), and Memory Snapshots (7-day TTL).
Buried in Modal's sandbox pricing FAQ is a crucial billing detail: because CPU Sandboxes are non-preemptible, they are billed at 3x base Modal Function CPU and memory list rates ($0.07074/vCPU-hr and $0.023976/GiB-hr).

Modal's saving grace for LLM agents is its elastic floor (max(request, actual)). On the default 0.125 physical core / 128 MiB floor, a sandbox waiting on LLM tokens costs only $0.0207/hr ($4.44 across 1,000 five-minute 85%-idle agent runs). If you hard-reserve 1 core / 2 GiB, idle cost jumps 9.2x to $0.1894/hr ($19.78 per 1,000 runs).
modal.com/signup offers zero email/password signup (OAuth only) and loads invisible hCaptcha.
Modal is the strongest pick when your coding agents or RL evals need massive parallel bursts, GPUs, or Docker-in-Sandbox microVMs. Just stick to elastic bursting instead of hard-reserving cores so the 3x non-preemptible multiplier does not wreck your idle-wait bill.
Software engineer and founder (15+ years building web apps). Every listing is verified against live product checks, official pricing pages, and real usage limits before publication.
Updated today