ARTICLE / SEPTEMBER 17, 2026
Most affordable AI agent sandbox cloud: workload costs
Compare AI agent sandbox costs using official pricing, explicit workload assumptions, and examples covering compute, idle time, storage, and plan fees.
By Platinum Blog
Pricing checked: September 17, 2026.
The most affordable AI agent sandbox cloud for your workload is the eligible service with the lowest cost after runtime, resource usage, retained storage, and plan fees are counted. Start with a workload budget, not a provider ranking: Platinum, E2B, and Fly.io Sprites publish different resource units and billing models. The calculations below use their official Platinum rates, E2B pricing, and Fly.io Sprites rates, not performance benchmarks.
Quick answer
Platinum is a sandbox cloud for AI agents that describes its sandboxes as microVMs with stop, resume, snapshot, and fork capabilities. At Platinum's published rates, 100 running sandbox-hours at 2 vCPU and 4 GB RAM cost $10.80 in compute, before disk and other excluded charges. Platinum charges for running compute rather than measured CPU utilization; reducing activity without stopping the sandbox does not reduce this modeled bill. These are vendor-described capabilities and list-price arithmetic, not independently tested performance claims. Source: Platinum product and billing FAQ.
Disclosure: Platinum Blog is Platinum's own publication, not an independent comparison site. Our publication and ownership introduction explains that relationship. This guide does not establish a universal cheapest provider: the same number of elapsed hours can produce different bills under allocated-resource and actual-use metering.
Comparison methodology
Normalize units, runtime, and utilization
For a reproducible budget, define aggregate sandbox-hours as the sum of running time across all sandboxes. The main Platinum example uses 100 sandbox-hours with 2 allocated vCPU and 4 allocated GB RAM throughout. Ten sandboxes running ten hours each therefore use the same modeled compute budget as one sandbox running 100 hours, but concurrency and session limits must be checked separately. Platinum's formula follows its published per-second running-compute rates. Source: Platinum billing FAQ.
Keep provider units visible. Platinum labels memory GB, while E2B labels memory GiB; the E2B illustration below uses 4 GiB, not an assumed equivalent of Platinum's 4 GB. Sprites prices cumulative CPU usage and actual memory usage, not the allocation-based Platinum example. We do not equate CPU-hours with allocated vCPU-hours, infer equal performance, or calculate a percentage saving between these examples. Sources: Platinum, E2B, Fly.io.
Include plans, credits, concurrency, storage, and egress
Separate recurring list-price budgets from introductory credits. Platinum describes prepaid credits with no subscription and a $10 new-account credit. E2B lists a $0/month Hobby plan with a one-time $100 credit and a $150/month Pro plan plus usage; upgrading to Pro does not add credits. The examples exclude credits rather than treating a temporary subsidy as the recurring cost. Sources: Platinum FAQ, E2B billing and limits.
Stopping compute does not imply the same storage bill everywhere. Platinum says a stopped sandbox costs at most its disk; E2B says billing stops when a sandbox is paused, killed, or times out. Fly.io lists Sprites hot storage while awake and cold storage for as long as data is retained. Budget storage across its billed lifetime, including stopped periods where applicable, rather than counting only execution hours. Sources: Platinum, E2B billing, Sprites pricing.
Official pricing table
All entries were checked September 17, 2026. Dollar figures use the providers' published dollar pricing. “Unknown” means not established by the cited evidence, not free or unlimited. The table covers selected products, not the entire sandbox market.
| Provider/product and official source | CPU, RAM, and disk rates | Billing basis | Plan/base fee and limits | Example subtotal and exclusions |
|---|---|---|---|---|
| Platinum | $0.0324/vCPU-hour; $0.0108/GB RAM-hour; $0.0526/GB disk-month | Compute per second while running; stopped disk may remain billable | Prepaid, no subscription. Vendor says hundreds concurrent per organization; exact allowance and session ceiling unknown. Default maximum shape: 16 vCPU, 32 GB RAM, 100 GB disk | 100 hours × 2 vCPU/4 GB: $10.80 compute. Disk, egress, taxes, and credits excluded; full invoice unknown |
| E2B pricing and billing limits | $0.000014/vCPU-second; $0.0000045/GiB RAM-second. Included storage: 10 GiB Hobby, 20 GiB Pro | Provisioned CPU/RAM per running second; published included storage has no extra charge | Hobby $0/month: 20 concurrent, 1-hour continuous runtime. Pro $150/month: 100 concurrent, 24-hour continuous runtime; paid concurrency upgrades available | 100 hours × 2 vCPU/4 GiB: $16.56 usage; $166.56 with a full Pro monthly fee. Not a 4 GB-equivalent comparison. Egress, taxes, credits, add-ons, and negotiated terms excluded; full invoice unknown |
| Fly.io Sprites | $0.07/CPU-hour; $0.04375/GB-hour actual memory; hot disk $0.000683/GB-hour; cold disk $0.000027/GB-hour | Cumulative CPU usage measured by cpu.stat; actual memory usage. Hot disk while awake; cold disk until deleted | No plans, tiers, or per-Sprite fee stated. Concurrency/session ceilings unknown from this pricing source | Allocation-only total unknown: requires measured CPU, memory, and storage. Illustrative usage calculations below exclude disk, egress, taxes, and credits |
E2B's official sources disagree about default RAM: the pricing representation describes 2 vCPU/4 GiB, while the billing guide describes 2 vCPU/512 MiB. E2B's pricing page places custom CPU/RAM under Pro, while its billing guide describes template-based resource customization and Hobby resource ceilings. The $166.56 example conservatively includes Pro; do not assume a customized 4 GiB workload qualifies for Hobby without checking the account configuration. If the selected workload is eligible for Hobby and fits its limits, the modeled usage is $16.56 with no monthly base fee, before credits and exclusions. Sources: E2B pricing, E2B billing.
Worked Platinum scenarios
100 running hours at 2 vCPU and 4 GB RAM
Platinum's compute-only estimate is 100 × [(2 × $0.0324) + (4 × $0.0108)] = $10.80. CPU contributes $6.48 and RAM contributes $4.32. The formula assumes the selected allocation remains constant for all 100 running hours; no storage allowance or introductory credit has been subtracted. Source rates: Platinum.
For an illustrative 10 GB of disk billed for one full month, add 10 × $0.0526 = $0.526, producing $11.326, or approximately $11.33, for compute plus that assumed disk charge. The disk-month is an explicit budgeting assumption, not a claim that 100 hours of compute implies a full month of storage. Platinum's FAQ says stopped sandboxes cost at most disk; confirm retention and proration for your lifecycle before treating this subtotal as a budget. Egress, taxes, and credits remain excluded. Source: Platinum rates and FAQ.
Stop the sandbox instead of merely leaving it idle
For the same Platinum 2 vCPU/4 GB allocation, reducing running time from 100 hours to 10 hours reduces modeled compute from $10.80 to $1.08. This scenario requires actually stopping the sandbox for the remaining time. A sandbox that stays running for 100 hours is still modeled at $10.80 even if its CPU is mostly waiting; retained disk remains a separate item. Source: Platinum running-time billing.
When other billing models may cost less
Sprites is Fly.io's persistent Linux environment product; its official pricing meters cumulative CPU usage and actual memory usage. Consider an illustrative 100-hour awake period accumulating only 20 CPU-hours and 40 GB-hours of memory usage: the compute subtotal is 20 × $0.07 + 40 × $0.04375 = $3.15. This is a deliberately low-use case, not a measured workload or a 4 GB-resident workload. Storage remains additional. Sources: Sprites overview, Fly.io pricing.
The Sprites memory assumption changes the result substantially. With the same 20 CPU-hours but 400 GB-hours of memory usage, the compute subtotal becomes $18.90; at 200 CPU-hours and 400 GB-hours it becomes $31.50. Actual-use billing can be attractive when measured use is low, but low CPU use alone does not establish a low bill. Neither illustration proves performance equivalence with Platinum or E2B. Source rates and metering: Fly.io.
E2B's included storage and no-billing paused state may matter for workflows that preserve many environments between short runs. E2B's plan fee and continuous-runtime limits still belong in the budget; a short evaluation using introductory credits is a different purchasing question from ongoing monthly operation. Model both periods separately rather than extrapolating the credited first bill. Sources: E2B pricing, billing and paused retention.
Choosing your sandbox
Use a small representative workload before committing to a provider. Record required resource allocation, actual CPU and memory usage where available, running versus stopped time, peak concurrency, longest uninterrupted job, retained disk, and outbound data. Then request any missing rate or limit and build two budgets: one at published recurring rates, another applying the credits or contract terms your account actually qualifies for. The differing billing inputs in the Platinum, E2B, and Sprites sources explain why a single headline hourly price is insufficient.
If the running-time model fits your application, review current Platinum pricing and configuration limits and substitute your sandbox-hours into the formula above. Treat the resulting number as a scoped estimate until retained storage and other invoice items are resolved. No example here promises a migration saving, faster execution, or a lower total invoice.
FAQ
Which AI agent sandbox cloud is cheapest for my workload?
The cheapest eligible provider is the one with the lowest total for your required allocation, actual usage, runtime, retention, and plan limits. This guide does not establish a universal winner. Platinum's $10.80 compute example, E2B's GiB-based estimate, and Sprites' actual-use examples use explicitly different inputs; they are budgeting illustrations, not equivalent-performance rankings. Sources: Platinum, E2B, Sprites.
Do idle AI agent sandboxes still cost money?
An idle but running Platinum sandbox remains subject to running-compute billing; stopping it can leave disk charges. E2B says billing stops when paused, killed, or timed out. Sprites distinguishes compute usage, hot storage while awake, and cold storage until deletion. Check the lifecycle state and storage policy rather than assuming that an agent waiting for input is free. Sources: Platinum, E2B billing, Sprites pricing.
How do I calculate Platinum sandbox compute cost?
Multiply running hours by allocated vCPU times $0.0324 plus allocated GB RAM times $0.0108. For 100 hours at 2 vCPU and 4 GB RAM, the result is $10.80 in compute. Platinum states that compute is billed per second, so hours are a convenient calculation unit rather than an asserted hourly billing minimum. Source: Platinum billing FAQ.
What charges are excluded from these examples?
Compute-only examples exclude disk, egress, taxes, credits, and any separately applicable fees. The Platinum disk illustration adds only the stated disk-month assumption; the E2B Pro illustration adds only the published monthly plan fee to usage. Unknown rates are not assumed to be zero, and no subtotal is presented as a complete invoice. Sources for included rates: Platinum, E2B, Fly.io.
Sources and checked date
Official sources checked September 17, 2026:
- Platinum product page, pricing, and billing FAQ: Platinum rates and vendor-described lifecycle. The homepage's competitor comparison remains labeled “verified Jul 2026”; its rival figures were not used as current comparator evidence.
- E2B official pricing: CPU/RAM rates, included storage, and plans; read alongside E2B billing and limits for runtime, retention, and configuration rules.
- Fly.io official pricing, Sprites section: CPU, memory, hot-storage, and cold-storage metering. The Sprites pricing URL redirected here when checked.
- Sprites official overview: product definition and lifecycle context.
- Welcome to the Platinum Blog: first-party ownership disclosure.
Prices and terms can change after the checked date. All workload totals in this guide are calculations from the linked rates, not observed customer bills or independent benchmarks.