Daytona Pricing 2026: Rates, Tiers & Real Monthly Costs

Do not index

Daytona pricing is pure usage-based: $0.0504 per vCPU-hour, $0.0162 per GiB-hour of RAM, and $0.000108 per GiB-hour of disk after the first 5 GiB free, billed per second, with $200 in free compute at signup and no subscription fee. Keep a standard 2 vCPU, 4 GB sandbox running around the clock and that works out to $120.89 for a 730-hour month, mid-pack among sandbox platforms and about 25 times the $4.76 that Agent37 (our product) charges for the same always-on shape. And to be precise about which Daytona: this guide covers daytona.io, the sandbox platform for running AI-generated code, not the watch, the race, or the beach.

We verified every number here against Daytona's own pricing page and docs on October 15, 2026. That date matters, because Daytona's pricing page publishes three unit rates and a GPU table and almost nothing else: no calculator, no example shapes, no monthly math. The third-party pages that rank for Daytona pricing queries mostly quote a single GPU rate or one sandbox shape with none of the lifecycle context that decides what you actually pay. This guide fills in the rest: how billing states work, the tier ladder that is not on the pricing page at all, and the worked math for real workloads.

Daytona pricing at a glance

The whole published rate card, from daytona.io/pricing:

Meter	Rate
vCPU	$0.0504 per hour
Memory	$0.0162 per GiB-hour
Storage	$0.000108 per GiB-hour, after the first 5 GiB free
Windows OS	$0.0858 per vCPU-hour, added on top of compute
GPUs	$0.57 to $6.25 per GPU-hour, by type (full table below)

"All billing is calculated per second," and there is no platform fee or required plan. New accounts get "$200 in free compute included" with no credit card required. Two things about that credit worth knowing before you budget around it: Daytona spends free credits before paid ones, and free credits cannot be used for GPU sandboxes at all. Startups can apply for up to $50k in free credits, and Enterprise (SSO, audit logs, BYOC, custom limits) is contact-sales.

One structural point separates Daytona from "pay for what you use" in the serverless sense: you are billed on reserved resources, not consumed ones. The billing docs say it plainly: "you are charged based on the resources reserved for your sandboxes." A sandbox provisioned with 4 vCPUs that sits idle bills all 4 vCPUs. The lever you control is state, which is the next section.

How Daytona billing actually works: states and timers

What a sandbox costs depends on its lifecycle state. From Daytona's billing docs:

State	What you pay
Running	vCPU + RAM + disk, all reserved resources
Creating, starting, stopping, pausing	Same as running, until the transition completes
Stopped	Reserved disk only
Paused (VM classes only)	Reserved disk only; the preserved memory state is free
Archived (container sandboxes only)	Nothing
Deleted	Nothing, but snapshots created from the sandbox keep billing for storage

Three timers move sandboxes between those states automatically. Container and GPU sandboxes auto-stop after 15 minutes of inactivity by default. VM sandboxes instead default to auto-pause after 60 minutes. A stopped container sandbox auto-archives after 7 days by default, which means an abandoned sandbox bills up to a week of disk before it goes free. Auto-delete is off unless you turn it on.

The auto-stop default hides the sharpest edge in Daytona billing: activity means external interaction, not CPU work. Daytona's own docs warn that "the sandbox may auto-stop mid-process because the process itself doesn't count as 'activity'." A 20-minute test suite or inference job with no incoming requests can be stopped at minute 15. The fix is setting the auto-stop interval to 0, but then a forgotten sandbox bills full rate indefinitely. Budget for one of those two outcomes.

Two more fine-print items from the billing docs. Transition states bill at the full running rate: "reduced billing applies once the sandbox reaches the stopped or paused state, not when the stop or pause operation begins." And charges can trail your usage by up to 48 hours, so usage that already happened can still post after you cancel; Daytona's docs make you responsible for deleting all sandboxes before cancelling.

Default sandbox resources are 1 vCPU, 1 GB RAM, and 3 GiB disk, and organizations cap at 4 vCPUs, 8 GB RAM, and 10 GiB disk per sandbox unless you request more.

The tier ladder: where Daytona pricing gets expensive

Nothing on the pricing page mentions tiers, but they decide what your money can actually do. Daytona places every organization into a tier that gates total compute and, more importantly, network access:

Tier	Org-wide pool	How you unlock it
Tier 1	10 vCPU, 30 GiB storage	Verified email
Tier 2	100 vCPU, 200 GiB RAM, 300 GiB storage	Credit card plus a $25 top-up
Tier 3	250 vCPU, 500 GiB RAM, 2,000 GiB storage	$500 top-up
Tier 4	500 vCPU, 1,000 GiB RAM, 5,000 GiB storage	$2,000 top-up every 30 days

(Daytona's docs list Tier 1 RAM as 10 GiB in one table and 20 GiB in another on the same page, so we are leaving it out.)

The network gate is the part that surprises people. On Tiers 1 and 2, "network access is restricted and cannot be overridden at the sandbox level." Essential services like package registries, git hosts, and model providers stay reachable, but general internet access starts at Tier 3, which costs a $500 top-up. One Hacker News commenter put it bluntly in October 2026 (on our own launch thread, so weigh the venue): "Got a bunch of Daytona credits which are close to useless given their restrictions until you hit tier 3."

So the honest minimum for an agent workload that browses the open web is not $200 of free credit. It is a $500 deposit. And note Tier 4 is not a one-time unlock: it requires a $2,000 top-up every 30 days to keep.

GPU workloads sit partly outside the ladder: on-demand GPU sandboxes count against a separate GPU quota and do not require Tier 3, while spot GPU sandboxes need Tier 2 or higher.

Daytona GPU pricing

Daytona sells eight GPU types in two flavors, preemptible (spot) and on-demand. The on-demand column lives in a second tab of the pricing page that is easy to miss, which is why third-party pages so often quote a single out-of-context rate. Here is the full table:

GPU	Preemptible	On-demand
Nvidia B300	$4.08/hr	$6.25/hr
Nvidia B200	$3.59/hr	$6.25/hr
AMD MI355X	$3.44/hr	$5.99/hr
Nvidia H200	$2.61/hr	$4.54/hr
Nvidia H100	$2.27/hr	$3.95/hr
Nvidia RTX PRO 6000	$1.74/hr	$3.03/hr
Nvidia RTX 5090	$0.74/hr	$1.29/hr
Nvidia RTX 4090	$0.57/hr	$0.99/hr

The caveats: spot sandboxes "can be terminated at any time without notice," with no preemption warning. GPU sandboxes are ephemeral, meaning they are deleted when stopped, so there is no cheap stopped state to park them in. Each GPU adds up to 16 vCPUs, 192 GB RAM, and 512 GB disk of capacity. And the $200 free credit does not apply to any of this; GPU time is paid-credit only.

What a month on Daytona actually costs

Daytona publishes no worked examples, so here is the math at the verified rates.

Always-on agent, 2 vCPU + 4 GB RAM: $0.1008/hr of vCPU plus $0.0648/hr of RAM is $0.1656/hr, which is $120.89 over a 730-hour month. Disk adds nothing if you stay within the first free 5 GiB.

The default sandbox, 1 vCPU + 1 GB: $0.0666/hr, or $48.62 per month if it never stops.

A bursty workload, 2 vCPU + 4 GB running 2 hours a day: about 61 running hours is roughly $10.07 a month, plus a stopped-state disk charge that rounds to zero at the 3 GiB default. This is the shape Daytona's model is built for: per-second billing, sub-90ms creation, and a disk-only stopped state make intermittent sandboxes genuinely cheap.

The same 2 vCPU + 4 GB shape on Windows: the $0.0858 per vCPU-hour OS surcharge adds $0.1716/hr, taking the month to $246.16.

Parked sandboxes: a stopped container with the maximum 10 GiB disk bills about $0.39 a month, and once archived it bills nothing.

Daytona pricing vs other sandbox platforms

Same workload, 2 vCPU + 4 GB RAM running 730 hours, every rate checked against the provider's own pricing page on October 15, 2026:

Platform	Unit pricing	Always-on month
Agent37	$0.80/vCPU + $0.70/GB RAM + $0.09/GB disk, per month	$4.76
Northflank	$0.01667/vCPU-hr + $0.00833/GB-hr	$48.66 (or a flat $48.00 plan)
Daytona	$0.0504/vCPU-hr + $0.0162/GiB-hr	$120.89
Modal Sandboxes	$0.00003942 per physical-core-second + $0.00000667/GiB-second	$173.71, or $143.71 after its $30 monthly credit; sandboxes cap at 24 hours
Cloudflare Sandboxes	$0.072/vCPU-hr of active CPU + $0.009/GiB-hr provisioned, on the $5 Workers Paid plan	~$165 at full CPU (closest shape carries 8 GiB RAM); ~$61 if the CPU mostly idles
E2B	Identical unit rates to Daytona, plus a $150/mo Pro plan for sessions over 1 hour	$270.89

Honest notes on that table. Modal's sandbox rates are three times its headline function rates, and it bills whichever is higher, requested or used. Cloudflare bills CPU only when it is active, so an agent that waits on an LLM most of the day lands far under the full-CPU figure. E2B's usage math matches Daytona exactly (the unit rates are identical to the digit), but keeping a sandbox alive past one hour requires the $150 Pro subscription, and even Pro sessions cap at 24 hours. Daytona has no session cap at all, which is a real advantage over both E2B and Modal.

For the full landscape beyond this one workload, see our 15-provider sandbox pricing guide and the Daytona alternatives comparison.

When Daytona pricing works for you, and when it does not

Daytona's model rewards a specific usage pattern. If your sandboxes are short-lived and bursty (CI-style code execution, test runs, scraping jobs that finish in minutes), per-second billing with a disk-only stopped state and free archival means you pay close to nothing between runs, and the $200 credit covers a long evaluation. No subscription fee means no floor while you experiment. That is a genuinely good deal, and it is why teams with intermittent workloads stay.

The economics invert when sandboxes accumulate real running time. An always-on agent pays the full $120.89 list rate per 2 vCPU + 4 GB shape, every month. Open internet access requires parking $500 in the wallet first. The default auto-stop will kill long unattended jobs at minute 15 unless you disable it, and disabling it converts forgotten sandboxes into full-rate meters. None of this is hidden once you read the docs, but none of it is on the pricing page either.

That always-on case is the one we built Agent37 for, so weigh the source accordingly. The same 2 vCPU, 4 GB sandbox with a 4 GB persistent disk costs $4.76 per 730-hour month, metered by the minute from a prepaid balance. A sleeping instance pays only its disk, about $0.36 a month, and wakes with files, memory, and installed tools intact. Isolation is documented by name (gVisor on standard instances, and full VM instances with their own kernel and KVM at 2.5x compute, $11.36 for the same shape). The pricing page compares more than 30 providers at published rates and backs it with a refund-the-difference guarantee: if you find a cheaper sandbox, we refund the gap. New eligible accounts get a $5 starter credit after adding a card (nothing is charged for adding it), which covers just over a month of the $4.76 entry sandbox.

If your agents run occasionally, Daytona's math is fair. If they run all day, run them on Agent37 Cloud and keep the other $116.

FAQ

How much does Daytona cost per month?

There is no monthly fee; you pay per second for reserved vCPU, RAM, and disk. A 2 vCPU, 4 GB sandbox running nonstop costs $120.89 over a 730-hour month at the October 2026 rates. The same sandbox stopped costs only its reserved disk, and archived it costs nothing.

Is Daytona free to use?

New accounts get $200 in free compute with no credit card required, and startups can apply for up to $50k in credits. There is no permanent free tier beyond the credit, and free credits cannot be spent on GPU sandboxes.

Does Daytona charge for stopped sandboxes?

Yes, for disk only. A stopped sandbox bills its reserved disk ($0.000108 per GiB-hour past the first free 5 GiB) until it is archived or deleted. Stopped container sandboxes auto-archive after 7 days by default, and archived sandboxes bill nothing.

What is Daytona Tier 3 and why does it matter?

Tiers gate what your organization can do. Full internet access from sandboxes starts at Tier 3, which requires a $500 top-up; on Tiers 1 and 2, network access is restricted and cannot be overridden per sandbox, with only essential services like package registries and git hosts reachable.

Does Daytona cost more on Windows?

Yes. Windows sandboxes add an OS surcharge of $0.0858 per vCPU-hour on top of normal compute rates, which takes a 2 vCPU, 4 GB shape from $120.89 to $246.16 for an always-on month.

Daytona Pricing in 2026: Real Rates, Tier Gates, and the Monthly Math