An honest comparison

Plenty of products keep working after you leave. Few tell you what it cost.

Every fact in the table below is taken from the vendor's own published pages on September 1, 2026, and linked next to the claim. Where a vendor does not publish something, the cell reads "Not stated". It is never scored as absent.

The four things compared

Four properties, chosen because they can be checked.

Model quality, speed and taste are hard to compare fairly and change every few weeks. These four properties are structural. They are either published or they are not, and they decide whether work survives the moment you stop watching it.

Dedicated machine

  • One container per customer, with its own disk, database and address, rather than a sandbox allocated per task and reclaimed afterwards.
  • Marked partial where a vendor gives an isolated environment that is torn down between sessions.

Keeps working after you leave

  • Whether a job continues once the browser tab is closed, and whether the result and its history are waiting when you return.
  • Covered in the table by the combination of a dedicated machine and standing orders, since almost no vendor states it plainly on its own.

Standing orders

  • Recurring work on a schedule the customer sets, running without a person present, with a stated limit on how many can exist.
  • Marked partial where scheduling exists but the smallest interval or the number allowed is restricted.

Per-job cost shown

  • Whether the product reports what an individual piece of work cost after it ran, in units that can be checked against a published rate.
  • Marked partial where only a range, a credit count, or a category estimate is shown rather than the actual settled figure.
The table

Eight products, sourced to the pages their vendors publish.

Entry price is the cheapest paid tier a vendor lists publicly, not the tier that matches Second Machine feature for feature. Usage unit is the thing that gets counted down as work happens.

ProductEntry priceUsage unitDedicated machineStanding ordersPer-job cost shown
Second Machine$29.95 / moCredits, stated per planYes, one container per customerYes, 1 / 5 / 20In your monthly statement
ChatGPTopenai.com/chatgpt/pricing · scheduled tasks help$20 Plus · $200 ProMessage tiersNot statedYes, hourly at most; 5 tasks on PlusNo
Claude (Pro / Max)claude.com/pricing · Claude Code on the web$17 annual · $100–$200 MaxRolling 5-hour limitsPer-session VM, reclaimed when idleYes (Routines)No
Perplexity Computerperplexity.ai/hub/pricing · credits help$20 · $200Credits, 100 = $1Isolated per task, not dedicatedYes, hourly floor, 15 per threadRanges per task type
Manushelp.manus.im$20 · $40CreditsNot statedYes, 20 on ProCredit history per task
Genspark Clawgenspark.ai helpcenterNot statedCreditsYes, dedicated cloud computerYesNo
Maritimemaritime.sh/pricing · how it works$20 for 20 agentsAgents; nothing meteredYes, micro-VM per agentYes, cronNo, deliberately flat
Hosted OpenClaw (ClawCloud)clawcloud.sh/pricing$29 · $49 · $109Server specs; BYO keysYes, one instance per serverNot statedNo
Hostinger Managed OpenClawhostinger.com/managed-openclaw$5.99 intro → $11.99Compute; credits includedYes, isolated instanceQueues on scheduleNo

Prices and features change. Tell us if a row is out of date and we will fix it and say when we did.

Reading the colours

Green is a published yes. Amber is a partial: the property exists but with a limit, or the figure is a range rather than a settled amount. Grey is either a published no or a cell the vendor does not address.

Why prices are not like for like

A $20 message tier and a $29.95 machine plan buy different things. The table lists the entry price so the shape of each offer is visible, not so the cheapest number wins.

What is left out

Model quality, answer accuracy, app integrations and support response times are all absent, because none of them can be sourced to a single vendor page and checked by a reader.

Where other products win

What Second Machine deliberately does not do better.

A comparison table that flatters the product publishing it is worth nothing. Here are the places where a reader should pick something else, stated as plainly as the table above.

01 / Price per agent

Maritime is cheaper, and flat

Maritime lists $20 for twenty agents with nothing metered, which works out to a dollar per agent per month against $29.95 for one machine here. For anyone who wants many small always-on agents and does not care what each run cost, that is the better deal.

02 / Predictability

Flat rate beats metering for some

Metering is only an advantage if a reader wants to see the cost. A flat rate with no ceiling to hit and no top-up to configure is simpler, and Maritime chose that deliberately rather than by omission.

03 / Model choice

ChatGPT and Claude have far larger ecosystems

Both ship frontier models, mobile and desktop apps, voice, image generation, connectors to dozens of third-party tools, and an extension ecosystem. Second Machine runs open-weight models through a metered gateway and a small set of runtimes. On raw model capability and breadth of integrations, it is not close.

04 / Maturity

Bigger vendors have more people behind them

Support hours, compliance paperwork, uptime history and regional data residency are all areas where a large vendor is ahead. Anyone whose procurement process asks for those should ask for them before choosing on price or receipts.

Second Machine is the better choice on exactly two axes: a container that is yours alone and stays yours, and a flat price where the only thing that changes between plans is how many credits you get. Everything else on this page is someone else's strength.

How we keep this honest

Every cell is sourced, dated, and correctable.

Every competitor cell on this page comes from the vendor's own published page, linked inline next to the claim rather than collected in a footnote. Nothing here is taken from a review site, a press summary, or another comparison table. The whole page was checked against those pages on 2026-09-01, and that date is printed in the footer of every page on this site.

SourcedEach row links the vendor page it was read from. Where two pages were needed, both are linked.
DatedChecked 2026-09-01. Vendor pricing changes often, so a row is only as good as its date.
Not statedA property a vendor does not publish reads "Not stated". It is never written up as a no, because silence is not evidence.
CorrectableReaders can report a stale row. Send the vendor page that contradicts a cell and it will be corrected, with the correction date shown on the row.
No paraphraseWhere a vendor states a limit in its own words, the cell uses that wording rather than a friendlier version of it.

If a cell on this page is wrong, that is a defect in the product, not a marketing detail. Report it the same way a bug would be reported.

A trial, a card authorization, and a machine of your own.

Answer one question about what you do. Your machine starts on it before you finish reading the confirmation, and proposes tomorrow morning's standing order. Plans from $29 a month.

See the plans 14 days · 1,000 credits · no charge until day 15