Cloud Dev Environments Compared: Codespaces vs Gitpod vs CodeSandbox

The "works on my machine" era is ending. Cloud development environments (CDEs) have moved from experimental luxury to mainstream infrastructure. Whether you are onboarding a new developer, reviewing a pull request, or coding from a tablet on a train, a CDE lets you spin up a consistent, ready-to-code workspace in seconds rather than hours.

But the landscape has shifted dramatically heading into 2026. Gitpod rebranded to Ona and pivoted toward AI agent orchestration. CodeSandbox doubled down on its microVM architecture. GitHub Codespaces deepened its integration with Copilot and the broader GitHub ecosystem. And DevPod emerged as a credible open-source, vendor-neutral alternative.

This comparison cuts through the marketing to examine pricing, features, performance, and practical use cases for each platform based on publicly available data and documented developer experiences.

Pricing re-checked 2026-08-14. Every figure below was re-verified against the vendor's own pricing or billing page on that date, and the Ona (formerly Gitpod) section changed materially since this article first ran. The claim-to-source matrix at the end records exactly which numbers we could load from a primary page and which we could not.

What Effloow added: a claim-to-source matrix for CDE pricing#

CDE pricing is unusually easy to get wrong, because two of the four vendors here restructured their plans inside twelve months and the old numbers still rank in search results. Rather than restate figures that circulate in secondary coverage, we tried to load each number from the vendor's own billing page and recorded the result — including the one vendor whose pricing page we could not load at all. An uncited hourly rate is not a small problem when you are using it to forecast a team's monthly infrastructure bill.

The State of Cloud Dev Environments in 2026#

The CDE market looks fundamentally different from two years ago. The biggest single change: Gitpod Classic pay-as-you-go shut down on October 15, 2025. The sunset notice is blunt about the consequence — pay-as-you-go users cannot log in or start new environments after that date. Gitpod rebranded as Ona in September 2025 and pivoted its mission from IDE-centric workflows to AI-driven software engineering. CEO Johannes Landgraf opened the announcement with "IDEs defined the last era. Agents define the next." Ona positions itself as "mission control for your personal team of software engineering agents," and reported that in one week of internal usage its agents co-authored 60 percent of PRs merged on main and 72 percent of lines of code. Read that figure for what it is: a single week, self-reported, on the vendor's own codebase.

For teams that previously relied on Gitpod's managed SaaS cloud IDE, this was a disruptive shift. Gitpod Flex (now simply called "Gitpod" under the Ona umbrella) replaced Classic, but it is self-hosted only and was AWS-only at launch. Teams that want a pure managed cloud IDE from Gitpod no longer have that option in the way they once did.

Meanwhile, GitHub Codespaces has become the default CDE for teams already on GitHub, with deep Copilot integration and a generous free tier. CodeSandbox has carved out a strong niche with its instant-clone microVM approach. And DevPod offers a completely free, open-source path using the same devcontainer standard that powers Codespaces.

Pricing Comparison#

Cost is often the deciding factor for teams evaluating CDEs. Each platform takes a different billing approach, so direct comparisons require normalizing to practical usage scenarios.

GitHub Codespaces Pricing#

Codespaces bills on two axes: compute time and storage. You only pay while a codespace is actively running -- suspended instances do not incur compute charges. Every rate in the table below comes from GitHub's Codespaces billing documentation and was re-confirmed there on 2026-08-14.

Storage is billed at $0.07 per GB per month.

The free tier is substantial: GitHub Free plan users get 120 core hours per month plus 15 GB of storage. GitHub's Codespaces product page spells out the conversion directly -- 120 core hours is "60 hours of run time on a 2 core codespace." GitHub Pro users receive 180 core hours and 20 GB. This is enough for moderate individual use without ever paying a cent.

CodeSandbox Pricing#

CodeSandbox uses a credit-based system. VM credits are priced at approximately $0.015 each, with credit consumption varying by VM size. The Free plan includes 400 credits per month, which translates to roughly 40 hours on a Nano VM.

One caveat you should carry into any budget you build from this table. CodeSandbox's pricing and docs pages sit behind a bot challenge that refused every automated request we made on 2026-08-14, including a real browser session. The figures below match CodeSandbox's own published plan text as indexed from its domain, but unlike the Codespaces and Ona numbers, we could not load the live page to confirm they are current. Treat them as a starting point and check codesandbox.io/pricing yourself before committing spend.

The Pro plan at $9 per month covers up to 20 collaborators in a workspace, making it one of the more affordable team options. Additional credits can be purchased through add-ons with bulk discounts. On-demand credits beyond the included allotment are billed at approximately $0.15 per hour.

Gitpod / Ona Pricing#

With the pivot to Ona, pricing has shifted to Ona Compute Units (OCUs). This is the section that changed most since this article first ran, and the stale version is still widely cached, so check the figures against ona.com/pricing rather than against search snippets. The table below reflects that page as loaded on 2026-08-14.

There is no longer a free tier. If you are working from an older comparison that lists a $0 Ona plan with $10 of credit and 3 parallel environments, that plan is gone -- the entry point is now the Core tier at $20 per month. The inactivity auto-delete window also moved from 3 days to 7.

The other correction worth making explicitly: Ona is not a self-hosted-only product. Core runs on Ona Cloud as a multi-tenant managed service, deployed in AWS. Only Enterprise is self-hosted in the customer's own VPC, managed by Ona, and that tier supports AWS or GCP. Enterprise customers get managed updates, scaling, and healing, with data never leaving their own account.

DevPod Pricing#

DevPod is completely free. There are no licensing fees, subscriptions, or per-seat charges. You provision your own infrastructure -- local Docker, a remote VM, or any cloud provider -- and DevPod orchestrates the devcontainer on top of it.

The vendor behind DevPod (Loft Labs, now vCluster Labs) claims it is typically 5 to 10 times cheaper than managed CDE services on comparable infrastructure, on the reasoning that it provisions bare VMs directly rather than adding a managed abstraction layer with markup. Flagging this one honestly: that is a vendor marketing figure with no published methodology behind it, and it is the weakest-sourced number in this article. The underlying mechanism is sound -- you pay raw compute rates -- but treat the specific multiplier as unverified.

Cost Scenario: One Developer, 40 Hours per Month#

This is where the Ona repricing bites hardest. For a single developer at light usage, three of the four platforms still cost nothing, but Ona now has a $20/month floor for the same workload that used to be free. If you are a solo developer or a small team evaluating on cost alone, that gap is the single most decision-relevant number in this article.

Feature Comparison#

Features vary significantly across these platforms, especially as each targets different segments of the developer workflow.

Startup Speed#

This is one of the starkest differentiators. CodeSandbox's microVM architecture allows it to spin up entire VMs, clone them, and restore snapshots within approximately 2 seconds. The number comes from CodeSandbox's own engineering write-up, How we clone a running VM in 2 seconds, which documents the memory-snapshot technique behind it. This near-instant startup transforms workflows like PR review, where you can click a link and land in a running environment almost immediately.

GitHub Codespaces takes longer -- typically 30 to 90 seconds for a cold start, though prebuilds can reduce this significantly by pre-warming images. For teams that create and destroy environments frequently, the startup time difference adds up.

Gitpod and DevPod startup times depend heavily on the underlying infrastructure and image complexity since both run on infrastructure you control.

Editor Experience#

GitHub Codespaces provides the most polished browser-based VS Code experience, which makes sense given that Microsoft owns both VS Code and GitHub. Extensions, settings sync, themes, debugging, and terminal access all work as expected. You can also connect from the desktop VS Code client via the Codespaces extension.

CodeSandbox offers its own browser-based VS Code fork with collaboration features baked in. Its standout feature is "follow mode," where team members can follow each other across files and edit simultaneously -- similar to Google Docs for code.

DevPod takes a different approach entirely: it creates the environment but lets you connect with your own local IDE. This means you get your exact personal setup -- custom keybindings, extensions, themes -- without any browser-based compromises. For developers who are particular about their editor configuration, this is a major advantage.

AI Integration#

AI features have become a key differentiator for CDEs in 2026. Each platform approaches AI integration differently.

GitHub Codespaces + Copilot#

Codespaces has the deepest AI integration of any CDE, thanks to its tight coupling with GitHub Copilot. In early 2026, GitHub removed the Copilot Workspace technical preview waitlist for all Pro, Business, and Enterprise subscribers.

Key AI capabilities within Codespaces include:

• Copilot code completion works natively with no additional setup required.
• Copilot Chat provides in-editor AI assistance for explaining code, generating tests, and debugging.
• Copilot CLI is now included in the default Codespaces image and offers four specialized agents -- Explore, Task, Plan, and Code-review -- that can run in parallel.
• Copilot Workspace acts as an agentic development environment: assign a bug or feature, and it analyzes your repository, creates a technical plan, writes code across multiple files, and runs tests before generating a pull request.
• Enterprise users can choose between multiple AI models for different reasoning tasks.

CodeSandbox AI#

CodeSandbox initially launched its own AI assistant called Boxy, which was deprecated in mid-2024. The platform now integrates Codeium for AI-powered code autocomplete, providing inline suggestions as you write. This covers basic AI assistance but does not match the depth of Copilot's multi-agent capabilities or agentic workflows.

Gitpod / Ona AI Agents#

Ona's entire pivot is built around AI. Rather than offering AI as a feature within a CDE, Ona positions the CDE as infrastructure for AI agents. The platform provides sandboxed environments where AI agents execute engineering tasks autonomously while maintaining VPC security and auditability. This is a fundamentally different model -- instead of you coding with AI assistance, AI agents code in managed environments with you overseeing the results.

DevPod AI#

DevPod has no native AI features. Since it connects to your local IDE, you use whatever AI tools your editor supports -- Copilot, Codeium, Cursor, or any other extension. This gives you maximum flexibility but requires separate setup.

Performance and Developer Experience#

Environment Consistency#

All four platforms solve the "works on my machine" problem, but through different mechanisms:

• Codespaces and DevPod both use the devcontainer.json specification, which is an open standard maintained by Microsoft. This means a project configured for Codespaces works with DevPod (and vice versa) without modification.
• CodeSandbox supports Dockerfiles and has added devcontainer compatibility. Its microVM approach means every branch and PR gets its own isolated environment, so switching contexts never requires stashing changes, running migrations, or reinstalling dependencies.
• Gitpod Flex adopted devcontainer.json support and deprecated the older .gitpod.yml format, which improves portability but required migration effort from existing Gitpod users.

Git Provider Lock-in#

This is a critical consideration. GitHub Codespaces only works with GitHub repositories. If your organization uses GitLab, Bitbucket, or a self-hosted Git server, Codespaces is not an option.

CodeSandbox, Gitpod/Ona, and DevPod all support multiple Git providers. DevPod works with any Git host since it just needs a clone URL.

Onboarding Impact#

CodeSandbox documents a notable case study with Adverity, where onboarding time decreased from 40 hours to 5 minutes after adopting the platform. While this is a single data point and results will vary, it illustrates the potential impact of CDEs on developer productivity. The core benefit -- a new contributor clicks a link and lands in a fully configured, running environment -- is consistent across all platforms.

Network Dependency#

Managed platforms (Codespaces, CodeSandbox) require a stable internet connection. If your connection drops, you lose access to your environment. DevPod with local Docker has no such limitation. Gitpod/Ona, being self-hosted, can theoretically be accessed over a private network without public internet.

The Open-Source Alternative: DevPod#

DevPod deserves special attention because it solves many of the same problems without vendor lock-in or recurring costs.

Built by vCluster Labs (formerly Loft Labs), DevPod is an open-source tool that creates reproducible developer environments using the devcontainer standard. Each environment runs in a separate container on whatever infrastructure you choose: local Docker, any reachable remote machine, or a VM in any public or private cloud. The project is MPL-2.0 licensed with roughly 15,000 GitHub stars.

Check the maintenance signal before you standardize on it. As of 2026-08-14, the loft-sh/devpod repository's last push was 2025-11-14 -- about nine months without a commit. The repo is not archived and the tool works, but a CDE is infrastructure your whole team depends on, and a nine-month gap is the kind of thing worth confirming for yourself rather than discovering after migration.

Why DevPod matters in 2026:

• Zero cost: No licensing fees or subscriptions. You only pay for the underlying compute.
• Provider agnostic: Switch between AWS, GCP, Azure, Hetzner, or localhost with a single command. No vendor lock-in.
• devcontainer compatible: Projects configured for Codespaces work with DevPod automatically.
• IDE agnostic: Connect with VS Code, JetBrains IDEs, or any editor that supports remote development.
• Automatic shutdown: Unused VMs are shut down automatically, keeping costs minimal.

The tradeoff is clear: DevPod requires more initial setup and infrastructure knowledge. There is no managed dashboard, no built-in collaboration, and no dedicated support team. For platform engineering teams comfortable with infrastructure, it is a compelling option. For individual developers or small teams without DevOps capacity, a managed platform will be less friction.

Who Should Use What#

Choose GitHub Codespaces If:#

• Your repositories live on GitHub and your team already pays for GitHub Pro, Team, or Enterprise.
• You want the deepest AI integration (Copilot, Copilot Workspace, Copilot CLI agents).
• You value a mature, managed service backed by Microsoft's infrastructure.
• You need prebuilds for large repositories where cold start time matters.
• You are comfortable with GitHub-only git provider support.

Choose CodeSandbox If:#

• Fast startup speed is your top priority -- the 2-second microVM spin-up is unmatched.
• You review many PRs and want instant, disposable environments per branch.
• Real-time collaboration (multiplayer editing) is a regular workflow for your team.
• You need multi-git-provider support with a managed platform.
• You prefer a credit-based pricing model with a low entry point ($9/month for a team of 20).

Choose Gitpod / Ona If:#

• You are interested in AI agent orchestration as a development model, not just AI-assisted coding.
• You want a managed multi-tenant service and can start at $20/month (Core), or you need self-hosting in your own VPC and can go to Enterprise.
• Your organization requires SSO, audit trails, and enterprise-grade access controls (Enterprise tier).
• You need unlimited parallel environments or large machines (up to 32 cores / 128 GB) or GPU support.
• You are on AWS for Core; Enterprise self-hosting supports AWS or GCP.

Skip Ona if you were looking for a free managed CDE. That option no longer exists, and the $20/month entry point makes Codespaces' and CodeSandbox's free tiers materially cheaper for light use.

Choose DevPod If:#

• You want zero vendor lock-in and zero recurring costs beyond raw compute.
• You work across multiple cloud providers or want the option to switch freely.
• You already use devcontainer.json and want the same standard to work everywhere.
• You prefer coding in your local IDE with the full power of your personal setup.
• You have the DevOps knowledge to provision and manage your own infrastructure.

Key Takeaways#

1. GitHub Codespaces is the default choice for GitHub-native teams. With 120 free core hours per month, deep Copilot integration, and a polished VS Code experience, it is the path of least resistance for teams already in the GitHub ecosystem. The main limitation is GitHub-only repository support.

2. CodeSandbox wins on speed and collaboration. Its 2-second microVM startup and built-in multiplayer editing make it the strongest option for teams that frequently create, share, and review environments. The $9/month Pro plan covering up to 20 collaborators is competitively priced.

3. Gitpod's pivot to Ona changes the calculus entirely. If you relied on Gitpod Classic's pay-as-you-go plan, it shut down on October 15, 2025 and you need a new one. A managed option does still exist -- Core, on multi-tenant Ona Cloud -- but it starts at $20/month, so the free path smaller teams used to take is gone. Self-hosting is now an Enterprise-tier feature, not the default. The AI agent orchestration angle is forward-looking but still maturing.

4. DevPod is the best option if cost and flexibility are paramount. It is free, open-source, works with any cloud provider and any IDE, and uses the same devcontainer standard as Codespaces. The tradeoff is setup effort and no managed infrastructure.

5. The devcontainer.json standard is the common thread. Codespaces, DevPod, and now Gitpod Flex all support it. Investing in devcontainer configuration for your projects provides portability across platforms and avoids lock-in to any single CDE vendor.

6. AI integration is now a differentiator, not a bonus. GitHub Copilot's deep presence in Codespaces sets the bar. Other platforms offer lighter AI features (CodeSandbox via Codeium) or a completely different AI-native model (Ona). Evaluate whether you need AI as an in-editor assistant or as an autonomous engineering agent -- the answer shapes which platform fits.

For most development teams in 2026, the choice comes down to ecosystem alignment. If you live on GitHub, start with Codespaces. If you need speed and collaboration across git providers, CodeSandbox deserves a serious look. If you want full control and zero lock-in, DevPod is the clear answer -- after you check its commit activity. And if you are building toward an AI-agent-driven engineering workflow, Ona's vision is worth evaluating, with the understanding that the free tier is gone and you are starting at $20 per month.

Claim-to-source matrix#

Which numbers in this article we could load from a vendor's own page on 2026-08-14, and which we could not.

Related reading#

• GitHub Copilot Agent Mode in JetBrains — the agent workflow that Codespaces' Copilot integration is built around, evaluated on its own.
• Cursor vs Windsurf vs Zed — if the editor matters more to you than where it runs, start here instead.
• E2B Sandbox for secure code execution — the adjacent problem of giving AI agents a disposable environment, which is where Ona is betting.
• Hetzner Cloud GPU servers — relevant if DevPod's bring-your-own-compute model appeals and you want cheaper raw VMs than AWS.