Daytona

What it is for

Daytona is a sandbox cloud for AI-generated code: your agent creates an isolated environment in tens of milliseconds, execs a command, and tears it down. Use it when the job is run untrusted agent code (and, if needed, a GPU), not when you want a Zap.

It is not open source as the hosted product. New accounts include $200 of free compute. Published CPU is $0.0504/vCPU-hour; preemptible H100 is $2.27/hour on the public menu.

What makes it unique: the cover is Daytona’s official pricing chrome — GPU rows (B300 $4.08/h, H100 $2.27/h, 4090 $0.57/h) plus memory / storage rates on the same panel.

Key features

• Instant sandboxes — create from the SDK (daytona.create); homepage claims sub-90ms start (daytona.io).
• Process / files / Git / LSP — exec with streaming output, CRUD the filesystem, run Git, and get language-server completions inside the box (daytona.io).
• Computer-use + code interpreter — the same infra menu lists isolated Python / shell and a desktop-shaped path for agents (daytona.io).
• GPU sandboxes — attach a published GPU (H100 / H200 / B200 / 4090, …) from the pricing panel (cover).
• Secrets outside the box — credentials stay in Daytona Secrets and are substituted at egress, not written into the sandbox (daytona.io).
• Snapshots + archive — snapshot a box to reuse later; archive or delete when you want the sandbox gone (Billing).

Who it's for

Best fit: agent platforms that will create/destroy sandboxes in code and might attach a GPU later.

Poor fit: researchers who want one published 8× H100 they SSH for a week (Lambda).

Pricing

Verified 2026-10-22 (Asia/Shanghai) from daytona.io/pricing (cover). All compute is per second. Re-check the Preemptible / On-demand toggle.

CPU / memory / disk

• vCPU — $0.0504/hour
• Memory — $0.0162/GiB-hour
• Storage — $0.000108/GiB-hour after the first 5 GiB free
• Windows — $0.0858/vCPU-hour

Preemptible GPUs (cover, $/hour) — B300 $4.08; B200 $3.59; AMD MI355X $3.44; H200 $2.61; H100 $2.27; RTX PRO 6000 $1.74; RTX 5090 $0.74; RTX 4090 $0.57.

New accounts: $200 free compute, no card on the “Start for free” path. Startup program copy goes up to $50k–$100k credits — apply separately. Enterprise (SSO, audit, BYOC) is sales.

Limitations / costs

• Stopped sandboxes still pay disk. Archive or delete if you want $0.
• GPU rows on the cover are preemptible. On-demand is a different toggle — copy the one you will launch.
• $200 is a grant, not a monthly allowance.
• Not E2B’s $150 Pro seat. Daytona is usage; E2B is a plan + usage.

Alternatives

• E2B — $0 Hobby + $100 credit if you only need CPU sandboxes.
• Modal — functions if you do not want a sandbox VM.
• Runpod — published Secure Cloud GPUs if the job is a Pod, not an agent box.

Official tutorial

Official Daytona channel video: Daytona Code Interpreter: Building a Model Context Protocol Server for Claude Desktop (oEmbed author: Daytona).

Written path: Pricing plus Billing.

Website

https://www.daytona.io

Repository

https://github.com/daytonaio/daytona

Source

Catalog row curated for AI Tools Nav. Cover is Daytona’s official GPU / CPU price panel (H100 $2.27/h, vCPU $0.0504/h). Tutorial: Code Interpreter on @daytonaio (oEmbed).

Last verified

2026-10-22 (Asia/Shanghai) editorial check.