Daytona Pricing Explained: Rates, Tiers, and Workload Costs

Daytona pricing is pay-as-you-go with no plan fee. A running sandbox costs $0.0504 per vCPU-hour and $0.0162 per GiB-hour of memory, billed per second, and disk costs $0.000108 per GiB-hour after the first 5 GiB. New accounts get $200 in free compute without a credit card. Below are the rate card, the tier limits, what typical workloads cost, and how Daytona compares with E2B, Modal and Beam.

Daytona pricing at a glance

Daytona sells compute by the resource rather than by plan. You pay for the vCPU, memory and disk a sandbox reserves, per second, from one wallet of prepaid credits. There is no monthly subscription. Enterprise contracts add larger limits, SSO, audit logs and BYOC.

The rates come from Daytona's pricing page, and the billing rules below come from its billing docs. Billing follows what a sandbox reserves, not what its code uses, so an idle sandbox with 4 vCPUs costs the same as a busy one until it stops.

Is Daytona free?

Daytona has a free trial rather than a free plan. Signing up needs no credit card and includes $200 in free compute, which Daytona spends before any paid credit. The wallet shows when free credits expire. Free credits can't pay for GPU sandboxes.

Startups can apply for up to $50,000 in credits through Daytona's startup program. Beyond the trial, the account stays on the lowest tier until you link a card and top up, and that tier limits what you can run, as the next section shows.

How Daytona tiers limit what you can run

Daytona has no plan fee, but tiers decide how much compute you get. Every organization has a shared pool of vCPU, memory and disk across all its running sandboxes, and the pool grows as you verify the account and prepay credit. The tier also sets API rate limits and internet access.

Daytona's limits docs list Tier 1 memory as 10 GiB in one table and 20 GiB in another, so check the dashboard before you plan around it. The top-ups are prepaid credit, not fees, and they count toward your usage.

Three tier rules affect cost planning:

• Internet access. On Tier 1 and Tier 2, outbound network access is restricted at the organization level, and a sandbox can't override it. Essential services such as GitHub, npm and PyPI stay reachable, and full internet access starts at Tier 3, according to the network limits docs. An agent that calls outside APIs or downloads from other hosts needs the $500 top-up.
• Disk quota. Running sandboxes and stopped container sandboxes count against the disk pool. With the 8 GiB disk of a medium sandbox, Tier 2's 300 GiB allows 37 at once, even though its CPU pool would fit 50.
• Per-sandbox size. New sandboxes get 1 vCPU, 1 GiB of memory and 3 GiB of disk by default. A sandbox can have up to 4 vCPUs, 8 GiB of memory and 10 GiB of disk unless Daytona raises the limit.

What Daytona sandboxes cost per hour

A Daytona sandbox costs between about $0.07 and $0.33 an hour, depending on size. Daytona's default snapshots come in three container sizes (daytona-small, daytona-medium and daytona-large), and matching snapshots exist for Linux VM sandboxes, which run their own kernel and add pause, fork and memory snapshots. The pricing page lists one vCPU and memory rate and no separate VM rate.

Each row adds vCPU, memory and disk above the free 5 GiB. The medium sandbox, for example, is 2 × $0.0504 + 4 × $0.0162 + 3 × $0.000108, or $0.1659 an hour.

Windows sandboxes cost more. The pricing page lists Windows at $0.0858 per vCPU-hour against $0.0504 for Linux, and the smallest Windows size comes with 4 GiB of memory and 30 GiB of disk.

Does Daytona charge for stopped sandboxes?

Yes, for disk only. A stopped Daytona sandbox stops paying for vCPU and memory but keeps paying for its reserved disk. An archived sandbox costs nothing, and a deleted one costs nothing except snapshots taken from it, which stay billed as storage.

Stopped disk is cheap. A stopped 10 GiB sandbox costs about $0.39 a month once the free 5 GiB comes off, and a medium one about $0.24. Pausing a VM sandbox keeps its memory without billing for it, so you pay only for disk.

Transitions bill at the running rate. Daytona charges creating, starting, stopping and pausing sandboxes as if they were running, and the lower rate starts only when the sandbox reaches the stopped or paused state.

The default timers and what they cost you

Daytona's lifecycle defaults keep idle costs low, and they can also interrupt long work. The sandbox docs set these intervals:

• Auto-stop stops a sandbox after 15 minutes of inactivity. A long process that gets no outside interaction doesn't count as activity, so a 30-minute job can stop halfway. Set the interval to 0 to turn it off.
• Auto-pause applies to VM sandboxes and pauses them after 60 minutes idle when neither timer is set.
• Auto-archive moves a stopped container sandbox to object storage after 7 days, which ends its disk charge. The longest interval is 30 days.
• Ephemeral sandboxes, and every GPU sandbox, are deleted when they stop, so they never accrue stopped-disk charges.

Daytona GPU pricing

Daytona GPU sandboxes cost from $0.57 per GPU-hour for a spot RTX 4090 to $6.25 for an on-demand B200 or B300, and the sandbox's vCPU and memory may be billed on top. Each sandbox takes up to 8 GPUs. GPU sandboxes are ephemeral and use a separate GPU quota, so they don't draw on the tier's CPU pool.

What a Daytona GPU sandbox costs with CPU and memory

The GPU rate may not be the whole bill. Daytona's billing docs list vCPU, memory and disk as billed for GPU sandboxes while they run, the same as for CPU sandboxes, and the pricing page doesn't say the GPU rate includes them. What that adds depends on how you create the sandbox:

Set CPU and memory yourself when you create a GPU sandbox. Left at the request defaults, an on-demand H100 costs $3.95 for the GPU plus about $3.97 for CPU, memory and disk, if Daytona bills them on top the way its billing docs describe.

Spot vs on-demand GPU sandboxes

Spot capacity is roughly 40% cheaper with one catch. Daytona can end a spot sandbox at any moment when an on-demand sandbox needs the GPU, and it sends no warning. Spot also needs Tier 2, and when no spot GPU is free the create call fails instead of waiting. The GPU sandbox docs cover both modes.

A GPU sandbox that stops is deleted, along with its filesystem. Write results to a mounted volume before the auto-stop timer runs out, or set the interval to 0 for long jobs.

What typical Daytona workloads cost

Short runs are cheap on Daytona, and the free $200 covers a lot of them. An always-on sandbox costs about $121 a month at the medium size. For parallel work, the tier you need tends to matter more than the rate.

A few notes on those rows:

• The batch row assumes the runs don't overlap much. Tier 1's 10 vCPUs fit five medium sandboxes at once on CPU, and memory may allow fewer.
• The eval row needs 100 vCPUs, which Tier 2 has, but 400 GiB of disk, which is over Tier 2's 300 GiB. Reaching Tier 3 takes a $500 top-up, which the eval then spends down.
• The always-on row only works if you disable auto-stop. Otherwise the sandbox stops after 15 idle minutes, and anything in memory is gone.
• The H100 row sets CPU and memory explicitly. At the request defaults of 16 vCPU and 192 GiB, the CPU, memory and disk would add $31.77 over the same 8 hours. GPU sandboxes use their own quota, so they don't need a higher tier, but free credits can't pay for them.

Daytona vs E2B, Modal and Beam pricing

Daytona and E2B charge the same CPU and memory rates to the cent: $0.0504 per vCPU-hour and $0.0162 per GiB-hour. Modal Sandboxes cost about $0.0710 per vCPU-hour. Beam prices sandboxes and VMs by CPU tier: its Performance tier, the default for new accounts, costs $0.05 per vCPU-hour, level with Daytona, and $0.01 per GiB-hour, about 38% less. The Shared and Spot tiers cost less again.

Modal bills sandboxes per physical core, and a core equals 2 vCPU, so the table halves its per-core rate. Beam's tiers apply to both sandboxes and VMs: Shared uses shared vCPUs and tops out at 8 vCPU and 16 GiB, and Spot runs on spare capacity that can be reclaimed with 30 seconds' notice. A GPU sandbox's CPU and memory bill at Beam's GPU-attached rates instead. Modal bills sandboxes at three times its standard CPU price because they run on non-preemptible capacity. Rates come from the E2B pricing page, the Modal pricing page and the Beam pricing page.

Daytona pricing vs E2B pricing

The rates match, so plan fees and limits decide it. E2B adds a $150 monthly Pro plan for sessions past an hour and more than 20 concurrent sandboxes, and concurrency add-ons cost $500 a month per 500 slots. Daytona charges no plan fee and asks for prepaid top-ups instead: $25 for Tier 2 and $500 for Tier 3.

On the eval row, Daytona costs $66.37 plus a $500 top-up the run partly spends. E2B costs $66.24 in usage plus $150 for Pro. Daytona adds GPUs and Windows, and E2B puts every sandbox in a Firecracker microVM, where Daytona's default sandbox is a container that shares the host kernel. Our breakdown of E2B pricing covers E2B's limits in detail.

Where Daytona comes out ahead

Daytona needs no plan to scale, its $200 trial credit is the largest of the four, and it's the only one with Windows sandboxes. On raw rates it no longer costs less than Beam for CPU-only code execution. The 1,000-run batch (2 vCPU and 4 GiB for 83.3 hours in total) costs $13.83 on Daytona. On Beam it costs $11.67 on the Performance tier, $3.67 on Shared or $1.50 on Spot, and Beam doesn't bill disk yet. Spot work waits for spare capacity and fails if none frees up within 20 minutes, so it suits batches that can retry.

Where Beam costs less

Beam also comes out ahead for persistent machines and for GPUs. Beam has two products for agent work. Sandboxes are short-lived containers that can attach a GPU, and VMs are persistent CPU microVMs with their own kernel, systemd, Docker and SSH.

Take the always-on row. On Daytona, a medium sandbox with 2 vCPUs, 4 GiB of memory and 8 GiB of disk (3 GiB of it billed) costs $0.166 an hour, or $121.12 over a 730-hour month. That also needs auto-stop set to 0, since the default stops an idle sandbox after 15 minutes. A Beam VM with the same 2 vCPUs and 4 GiB costs $0.14 an hour on the Performance tier, or $102.20 over 730 hours. On the Shared tier it's $0.044 an hour, or $32.12, and Beam doesn't bill the 50 GiB root disk yet. Pass --cpu-tier shared when you create the VM to pick Shared. Pausing it keeps RAM and running processes, and stopped or paused VMs accrue no compute charges. An idle timeout can pause the VM on its own:

beam vm new devbox --cpu 2 --memory 4096
beam vm update devbox --ttl 300 --idle-action pause --auto-resume

With auto-resume on, the next SDK call or request to a published URL wakes the VM. Our guide to microVMs and containers explains why long-running agents benefit from their own kernel.

GPUs are the second gap. An on-demand H100 sandbox on Daytona costs $3.95 an hour for the GPU, plus CPU and memory if they're billed on top: about $0.92 more at 8 vCPU and 32 GiB, or $3.97 more at the request defaults. Beam's on-demand H100 machines start at $1.83 an hour with 26 vCPU, 200 GB of RAM and 1 TB of NVMe included, and Beam's serverless RTX 4090 costs $0.69 an hour for the GPU. A Beam sandbox attaches a GPU with one argument through the same API, though requesting a GPU doesn't install CUDA libraries:

from beam import Sandbox

sb = Sandbox(gpu="RTX4090", memory="8Gi", keep_warm_seconds=300).create()
try:
    result = sb.process.run_code("import os; print(os.cpu_count())")
    print(result.stdout)
finally:
    sb.terminate()

Beam's plans work differently from Daytona's tiers. The Developer plan costs $20 a month and returns it as $20 of usage credit, and it allows 30 vCPUs and 5 GPUs running at once. That limit is the total CPU across VMs, sandboxes and serverless apps, not a count of containers, so it fits seven default 4-vCPU VMs or fifteen of the 2-vCPU VMs above. The Team plan costs $89 a month plus usage and allows 1,000 vCPUs and 50 GPUs at once. There are no egress or bandwidth fees on any plan. Our Daytona alternatives guide compares the wider field, including Northflank, Tensorlake and Fly.io Sprites.

FAQ

How much does Daytona cost?

Daytona charges $0.0504 per vCPU-hour, $0.0162 per GiB-hour of memory and $0.000108 per GiB-hour of disk past 5 GiB, billed per second with no monthly fee. A default 1 vCPU, 1 GiB sandbox costs about $0.07 an hour, and the largest standard size costs about $0.33.

Is Daytona free?

Daytona offers a free trial with $200 in compute and no credit card. It has no permanent free plan, and free credits can't pay for GPU sandboxes. Startups can apply for up to $50,000 in credits.

Does Daytona charge for stopped sandboxes?

Only for disk. A stopped sandbox pays $0.000108 per GiB-hour for reserved disk past 5 GiB, and nothing for vCPU or memory. Archived and deleted sandboxes cost nothing, and container sandboxes archive automatically after 7 days stopped.

Is Daytona cheaper than E2B?

Their CPU and memory rates are identical. Daytona has no plan fee, while E2B charges $150 a month for Pro, which you need for sessions over an hour or more than 20 concurrent sandboxes. Daytona instead asks for prepaid top-ups of $25 or $500 to unlock bigger pools and full internet access.

How much does a Daytona GPU sandbox cost?

GPU rates run from $0.99 an hour for an on-demand RTX 4090 to $6.25 for a B200 or B300, and spot capacity runs roughly 40% less. Daytona's billing docs also list the sandbox's vCPU and memory as billed, so set them explicitly: a GPU sandbox created with no resources set gets 16 vCPU and 192 GiB, which would add about $3.97 an hour. Spot sandboxes can be ended without warning and need Tier 2.

Does Daytona have a monthly fee?

No. Daytona is pay-as-you-go from a prepaid wallet. Higher tiers require one-time top-ups ($25, then $500), and Tier 4 requires a $2,000 top-up every 30 days, all of which become usage credit.

Price an agent workload on Beam

Beam runs short tasks in sandboxes that can attach a GPU, and long-running agents in persistent VMs that cost nothing for compute while paused. Create a Beam account and compare your own workload against the rates above.