Skip to content
Independent notes on AI computeChecked daily / 2026.09.26
buildai/current

Credits

Last checked:

Daytona free credits: $200 in free compute

creditssandboxdaytonagpu

A cloud sandbox platform for running AI-generated code, billed per second for CPU, memory, and storage. Official site

Daytona signup credit, the facts

Official pricing

Amount$200 in free compute
Grant typeOne-time signup credit
ExpiryNot stated officially
Payment method requiredNo
Withdrawable as cashNo
What I ran on itNot tested yet

Filed under: GPU cloud free creditsAI sandbox free creditsFree credits with no card required

Good for

  • Agent code execution in CPU sandboxes without entering a card
  • Per-second billed experiments where $200 goes a long way at $0.0504 per vCPU hour

Not for

  • GPU sandboxes on the free credit: the billing docs exclude them
  • A persistent server you administer: sandboxes are the product

Daytona’s pricing page says “$200 in free compute included” and “no credit card required.” It is a signup grant, not a monthly allowance, and the first 5 GiB of storage is free on top of it. How long the credit stays valid is not stated officially: the wallet UI has a “See breakdown and expiration” control, but no public page states a validity window. Checked against the official pricing, billing, limits, and sandboxes docs pages on 2026-08-24.

The catch: free credit does not pay for GPU sandboxes

This is the single most important line on the page. Daytona lists GPU sandboxes with per-hour rates, as of 2026-08-24: H200 at $2.61, H100 at $2.27, RTX PRO 6000 at $1.74, RTX 5090 at $0.74, and RTX 4090 at $0.57 per hour. But the billing docs state the free credit balance is “not available for GPU sandboxes.” So the $200 cannot buy a single GPU minute. If a blog tells you Daytona hands out $200 of free H100 time, that is wrong on both ends of the sentence. If you want a grant that can touch GPUs at all, the Modal free tier is a monthly allowance on a platform whose pricing includes GPUs, though Modal does not itemize coverage either.

What the $200 actually buys

CPU sandbox compute, billed per second. The pricing page lists vCPU at $0.0504 per hour, memory at $0.0162 per GiB hour, and storage at $0.000108 per GiB hour after the first free 5 GiB. Windows sandboxes cost $0.0858 per vCPU hour. At those rates the credit covers a lot of agent runtime, which is the same territory as the E2B Hobby credit, a one-time $100 with no card required. The two grants are side by side in E2B vs Daytona.

No credit card required

The official wording is “no credit card required,” and I mirror it exactly rather than extending it to payment methods in general. New accounts start in Tier 1 of the limits docs, which requires a verified email and caps resources at 10 vCPUs and 30 GiB of storage; the page’s two tables disagree on the memory cap, listing 10 GiB in one and 20 GiB in the other.

Billing gotchas beyond the credit

Two behaviors from the billing docs are worth knowing before you burn the grant. GPU sandboxes are auto-deleted when stopped, which matters only once you pay for GPUs anyway. Stopped non-GPU sandboxes still bill for disk, so a pile of stopped sandboxes quietly eats credit through storage charges.

The startup program is a separate thing

Daytona also advertises a startup program with up to $50,000 in credits. That is an application-based program, not the signup grant, so do not read the $50k figure as something every new account gets.

Auto-stop and auto-archive: the timers behind the $200

How fast the credit drains depends less on the hourly rates than on Daytona’s idle timers, documented in the sandboxes docs. For auto-stop, “if not set, the default interval of 15 minutes is used,” and setting 0 disables it. The docs add a sharp edge for agent workloads: “If you run a long-running task like LLM inference that takes more than 15 minutes to complete without any external interaction, the sandbox may auto-stop mid-process” because internal processes do not count as activity, so the idle timer never resets. VM sandboxes run on a different clock: when neither interval is set, classes that support pausing “default to an auto-pause interval of 60 minutes with auto-stop disabled.”

After a container sandbox stops, a second timer starts. The auto-archive interval moves a continuously stopped sandbox to object storage, with “the default interval of 7 days” and a special zero value: setting 0 means “the maximum interval of 30 days is used.” The billing docs map each state to a charge: a started sandbox bills for vCPU, memory, and disk; a stopped one bills for disk only, as noted above; an archived one is “Not billed.” Transitional states (Creating, Starting, Stopping, Pausing) are “Billed the same as started.” So with defaults, an abandoned sandbox burns full rate for up to 15 minutes, disk rate for 7 days, and nothing after archiving.

What I have done with it

Not tested yet. I have not personally run sandboxes against the $200, so this page reports only what the official pages state.

For how sandboxes compare with a VPS, GPU cloud, or serverless platform, read how to choose compute. Every grant I track is on the credits index.

What changed here

  1. Correction

    Corrected the Tier 1 limits sentence: the limits docs list 10 vCPUs (not 10 sandboxes) with email verification as the tier requirement (not a pre-verification state), and the page's two tables disagree on the memory cap (10 GiB vs 20 GiB).

  2. New entry

    Added Daytona: $200 in free compute.

Plus 1 recheck that found nothing changed. Every row links to the full record on the changelog, with the official page it was checked against.

Related entries

Sources

Last checked:

All numbers on this page come from the official pages above, checked on the date shown. If a figure is not stated there, this page says "Not stated officially" instead of guessing.