Battleships

The most comprehensive comparison of platforms offering cloud runtime / sandbox / VM compute for agents

Usecase preset:

Suggested features tap to require or drop

Scale same settings as the panel

    How the ranking shifts with volume

    Monthly bill for this workload scaled from 0.1× to 10× sessions (concurrency and egress scale with it). Plan fees, minimums and pool tiers make lines cross.

    Methodology

    One pipeline and one set of rules for every provider, boat.dev included.

    1. Read

      Agents open every pricing page and doc in a real browser.

      0 pages read
    2. Quote

      Every answer links to its source and quotes it word for word. No quote, no answer.

      0 quoted answers
    3. Price

      One engine bills your usage the way each provider does: per second, minimums, idle time, starts, plans, seats.

      0 pricing tiers
    4. Rank

      Full matches first, cheapest on top. The rest show what they miss, so you decide what you can live with.

      0 providers

    Rules we don’t bend

    • Unknown stays unknown. A price or limit a provider doesn’t publish is never read as free or unlimited.
    • Disclosure never costs you. Publishing a limit can’t rank worse than hiding it, and what’s true of everyone counts against no one.
    • List prices first. Promotions, spot capacity and sales deals only count when you allow them.
    • Built by boat.dev, not for it. Same engine and rules for boat.dev. The code and data are open, and we welcome contributions.

    What we can’t check

    • That every feature really works. We didn’t test each feature end to end, and providers sometimes claim things that don’t work.
    • Reliability. Beyond our benchmarks we can’t measure uptime or flakiness, and some status pages aren’t honest.
    • Sales and support. We can’t tell how a sales deal goes or if support answers email. Only self-serve is certain.
    • Company health. We can’t see customers or revenue. Funding and a solid customer base matter when you pick a provider.
    Every pricing page, read by an agent in a real browser inside boat.dev VMs, using the pi harness and Claude Haiku 4.5: 361 of them at once, at 4× speed.

    For AI agents: /llms.txt · Data and code: GitHub · Spotted a wrong price? Tell us