AI agents now work for hours, not seconds. The length of task a frontier model can finish on its own has been doubling at a steady pace for years, so the cost of an agent's virtual machine now depends on how a provider bills the long stretches when the agent is thinking, waiting on a model or sitting idle.
Sail Research built its pricing around that shift and launched from stealth with backing from Kleiner Perkins and Sequoia to serve long-running agents. Sail bills a Sailbox only for the CPU, memory and disk it actually uses, and nothing while it sleeps or pauses. Solari bills a sandbox by its size while it runs, at rates that fall with each plan, and puts cloud browsers and desktops on the same account. This guide covers what each charges for, how the two models compare and how to estimate your own bill.
What Does Sail Charge For?
Sail charges for measured use, not reserved capacity. Sail says a Sailbox pays per hour for the vCPU time, RAM and NVMe disk it actually uses, so a Sailbox sized for several vCPUs that uses a sliver of one pays for the sliver (Sailbox pricing). Sleeping, paused, checkpointing and cold-starting time is not billed.
Two smaller line items sit on top. Volume storage bills for every hour the volume exists, including while attached Sailboxes sleep, and each new Sailbox on Sail's Free plan carries a one-time creation charge, copies from checkpoints included. The Pro and Enterprise plans waive that charge.
Sail also sells open-model inference on the same account, paid per token. Its Free plan comes with monthly credits once a payment method is attached, and its Pro plan adds included credits, unlimited seats and a much larger allowance of reserved vCPUs. When credits run out, Sail pauses running and sleeping Sailboxes rather than deleting them.
Why Does the Billing Model Matter for Agents?
The billing model decides who pays for idle time. A long-running agent spends much of its session waiting on a model response, a slow build or a person reviewing its work. Under size-based billing a running machine costs the same busy or waiting; under usage-based billing a waiting machine costs close to nothing in CPU.
Memory behaves differently. Sail counts application memory, guest kernel memory and the Linux page cache toward billed RAM, so an agent that reads a large repository holds memory it pays for even when its own process is small. Memory keeps billing while CPU drops toward zero, so the memory rate is the first number to compare for an agent that mostly waits.
How Do Solari and Sail Prices Compare?
Solari bills compute by machine size. Solari charges $0.0245 per vCPU-hour and $0.0077 per GB-hour on the Professional plan, and $0.0175 and $0.0055 on Team. Sail publishes per-unit rates for used vCPU, used RAM and used disk; because Sail's unit is CPU time actually used while Solari's is the vCPU you size the machine with, the two line up directly only for a sandbox running flat out.
Solari's rates fall at every plan, and an Enterprise contract drops compute to $0.00175 per vCPU-hour and $0.00055 per GB-hour for fleets. For a busy agent, compare Solari's rate for the size you need against Sail's rates at full load; for a mostly idle agent, compare how each handles the waiting time.
How Do You Estimate Your Sandbox Bill?
You can estimate either bill from three things you already know: how big the machine is, how many hours it runs and how much of that time it works.
Measure busy and idle time. Log how long your agent computes against how long it waits on a model or a person. A high idle share favours usage billing on CPU; a busy agent brings the two models close together.
Measure real memory. Check resident memory and file cache during a typical run, not the machine limit. On Sail that figure drives the bill; on Solari, pick the smallest size that holds it.
Count forks and copies. Agents that branch work into many copies pay for every copy on both providers. On Solari, stored snapshots cost $0.05 per GB-month with the first 10 GB free, so you can keep checkpoints without watching the meter.
Add the browser. Many agents also need a real browser, driven through Playwright, to sign in, research or file a form. A second vendor for it means a second bill and a second integration.
Weigh the fee against credits. Solari's Starter and Professional plans cost $20 and $200 a month and return the same amount as usage credits, so the fee costs nothing extra once your usage passes it.
What Does It Cost on Solari?
Solari is the infrastructure an agent runs on: cloud browsers, sandboxes and desktops from one API and one bill. For long-running agents, these are what keep the bill down:
Pause instead of paying for idle. Solari sandboxes pause after an idle timeout by default. Pause saves memory and processes, and resume brings the sandbox back exactly as it was, so an agent that waits overnight stops running instead of holding a machine.
Snapshot, revert and fork. Take a snapshot of a running sandbox, then revert to it or fork it. Your agent can try three approaches from the same starting point and keep the one that works.
Low rates as you grow. Solari runs a 4 vCPU, 8 GB sandbox for $0.114 an hour on Team, and every rate drops again on an Enterprise contract, so the price per task keeps falling as your fleet grows.
Desktops for computer use. A Solari VM gives computer use models a full Linux screen, is ready "usually in about a second" and adds $0.02 an hour for the live screen. A paused VM stops counting against your running limit.
No overage surprises. When your balance reaches zero, new sandboxes and desktops stop launching, and Solari never sends an end-of-month overage charge.
Fleet scale. Solari supports up to 100,000 concurrent sandboxes or desktops on Enterprise, so you can run every task in a batch at once instead of queueing them.
Browsers sit on the same account at $0.07 an hour on Professional, so an agent that writes code and works the web runs on one key and one invoice.
Which Pricing Model Fits Your Agents?
Sail bills the CPU, memory and disk a Sailbox actually uses and skips idle time; Solari bills the size you choose while a sandbox runs, pauses it when it goes idle and lowers every rate as you move up a plan. Agents that mostly wait reward usage billing on CPU, while busy agents, large fleets and agents that also need a browser or a desktop favour Solari's rates and single account.
Over the next year, expect three changes in how agent infrastructure is priced:
Billing that tracks real use. More providers will charge for what an agent consumes rather than what it reserves, and buyers will compare memory rates as closely as CPU rates.
One bill for the agent's whole computer. Teams will buy the sandbox, the browser and the desktop together, because their agents use all three in one task.
Contracts for fleets. As agents run in the thousands, volume rates will decide the bill far more than list prices.
What stays human is the judgment about which tasks are worth an agent's hours, and who checks the work before it ships.
Get Started With Solari
Sail pairs its sandboxes with its own model inference and bills on measured use; Solari is the infrastructure specialist and moves faster on speed and price across browsers, sandboxes and desktops. Start with one long-running agent, measure its busy time and memory for a week, then price it on both models. Solari is the fastest and cheapest way to run cloud browsers and VMs for long-running agents. Get started with a free account and start running your agents today.
