What’s new in Playcode

In the editor's Preview, a link that opens a file in a new tab - a PDF from your public folder, an image, a download - now opens that file, the same way it does on your published site.

• Before: the new tab showed "The connection has been lost", which made

a correct link look like a bug in your site.

• Now: the file opens. Pages and routes behave exactly as before.
• Direct links to the preview (a bookmark, or a page link opened outside

the editor) now say what they are: the preview only runs inside the Playcode editor. Use Open preview in a new window from the preview toolbar to view your project in its own tab.

Search Workspace > Domains and the list just got a lot longer: 615 new extensions are now searchable and buyable, priced at what the registries actually charge today.

• Recognizable names, not just filler. .xyz, .zip, .yoga,

and .zone are in, alongside hundreds of others spanning tech, business, and lifestyle endings.

• Your originals are untouched. .com, .net, .org, .app,

.dev, and .io stay at the price they have always been - this expansion only adds, it never repriced what you already use.

• A few extensions still ask for more. Some endings - country codes and

a handful of membership-only names - need something extra from their own registry before anyone can register them through us. Search still tells you what is missing instead of a dead end; we are working through that list one registry at a time.

Apple shipped a major Expo Go update this week, which broke the phone preview for mobile projects: scanning the QR code showed an SDK-compatibility error on iPhone. Fixed on both sides - new mobile projects target the Expo version the App Store actually ships, and the AI now checks your project against the current Expo Go before handing you a QR code, upgrading the project first if it is out of date.

The QR code itself also got more reliable: it now always carries the secure exps:// link, so it opens in Expo Go on the first scan.

GPT-6 Astra - OpenAI's most capable model, built for the hardest end-to-end work - is now available on the Ultra tier, next to Claude Fable 5.1. OpenAI launched it yesterday behind a limited-access program; our access opened this morning and it went through our full verification the same hour: live tool-calling, prompt caching, and token accounting all confirmed on the real wire before it reached the picker.

Astra is priced identically to Claude Fable 5.1 ($10 / $50 per million tokens at the provider), and because its tokenizer encodes the same work to roughly a third fewer tokens, the same build tends to cost fewer credits. Claude Fable 5.1 remains the Ultra default.

Background agents (the tasks your AI delegates work to) can now look at images, capture page screenshots, and understand videos, just like the main agent. Visual work delegates reliably: captioning a photo gallery, writing honest alt text, matching a reference design, picking a video poster frame, or reviewing a page's layout no longer fails partway because the helper agent could not see what it was working with.

Available on vision-capable models across all task roles - explorer, coder, auditor, and researcher.

Gemini 3.8 Flash is now the Gemini option on the Pro quality tier. It launched today at the same price as its predecessors, and Google's own launch numbers put it ahead of 3.7 Flash on every published axis - software engineering, terminal coding, and multi-step reasoning. In our own SVG drawing benchmark it is measurably the strongest Gemini Flash yet.

It also costs you fewer credits than 3.6 Flash did. Google discounted this rate card after we first added 3.6 Flash, and our pricing had not caught up, so 3.6 Flash was charging roughly double what it should have. That is corrected: every Gemini Flash model now bills at the real rate.

If you had 3.6 Flash selected on a chat, it keeps working and stays on that chat - nothing you pinned has moved. It is simply no longer offered in the picker for new selections.

The cheap-and-fast tier is unchanged. Gemini 3 Flash is still the default there, and still the cheapest way to build.

Grok 4.6 is now the Grok option on the Pro quality tier, in place of Grok 4.5. xAI prices the two identically and calls 4.6 "the most intelligent and fastest model we've built" - and in our own SVG drawing benchmark the difference is visible: where 4.5 famously spent 668 tokens barely attempting the task, 4.6 actually engages with it.

If you had Grok 4.5 selected on a chat, it keeps working and stays on that chat - nothing you pinned has moved. It is simply no longer offered in the picker for new selections.

GLM 5.3 is now the GLM option on the cheap-and-fast tier, in place of GLM 5.2. Z.ai prices the two identically, and 5.3 is the newer and stronger model - Z.ai's own measurements show it doing markedly better agentic coding work while spending *fewer* output tokens than 5.2 at every reasoning-effort level, which on a reasoning model means the same work tends to cost you less.

It also brings a real reasoning-effort dial (low / high / max), which the Intelligence setting now drives.

If you had GLM 5.2 selected on a chat, it keeps working and stays on that chat - nothing you pinned has moved. It is simply no longer offered in the picker for new selections.

GPT-5.6 Luna is now available on the cheap-and-fast tier. OpenAI cut its price fivefold today - to $0.20 in / $1.20 out per million tokens - which makes Luna the cheapest model in the picker, below even Gemini 3 Flash. In our SVG drawing benchmark it lands right beside Gemini 3 Flash in capability, with a reasoning-effort dial that responds cleanly to the Intelligence setting.

We also passed today's OpenAI price cuts for GPT-5.6 Sol and GPT-5.6 Terra straight through - the same work now costs you fewer credits on all three.

Gemini 3 Flash remains the tier default.

You can now connect email services like MailerLite, SendGrid, Mailchimp, Klaviyo, and Postmark to a domain managed by Playcode.

These services verify your domain with DKIM records: CNAME records whose target contains an underscore, such as litesrv._domainkey.mlsend.com. Playcode's DNS editor rejected those targets as invalid, which blocked email authentication. Underscores are now accepted in CNAME, ALIAS, NS, and MX targets, so you can add the records exactly as your email service shows them.

The app now opens in your browser's language on your first visit - all ten languages, including the sign-in page. On the website, the language switcher now also lives in the top menu (next to Pricing on desktop, inside the menu on phones), and if the site isn't in your language, a small note offers the switch in your own language.

Also fixed on phones: on some Android screens the page could grow slightly wider than the display, which pushed the menu button half off the edge. The menu stays put now, and a new account button next to it takes you straight to sign-in - or to your projects if you're already signed in.

Claude Fable 5.1 is now the default model on the Ultra tier, replacing Claude Fable 5. It is Anthropic's most capable model, tuned for the hardest reasoning and long agent sessions - and it comes with a real cost improvement: reading cached context is 4x cheaper than on Fable 5. In practice that is most of what a long coding session pays for, so the deeper into a project you are, the more you save.

Token prices are unchanged. Claude Fable 5 remains available in the model picker if you have it selected.

Attaching files used to mean hitting a small target inside the chat box. Now the whole editor catches your files:

• Drop anywhere in the chat - the panel lights up while you drag, and

the file attaches to your message.

• Drop an image straight onto your page. Drag it over the preview and

the element under your cursor highlights - release, and the image lands in the composer aimed at that exact spot. Press Enter and the AI places it there.

• Drop on Home to start a new project with the file already attached.
• While you drag, the sidebar shows your project's file tree, so you

can also drop a file straight into a folder.

Every drop answers visibly: the file flies to where it landed, and if you drop somewhere that can't take files, a short note points you to the chat or the file tree. A missed drop never navigates the page away or loses what you typed.

Publishing asked the wrong question about who you are. If you were a workspace admin opening a project a teammate had created, the Publish button appeared and looked ready - but pressing it did nothing at all. No progress, no message, no explanation. Only the person who originally created the project could publish it.

The button was reading your team permissions, while the publish action behind it was still checking whether you personally created the project. Both now read the same rule: if you can manage the project, you can publish it. Unpublish was affected in the same way and is fixed too.

Nothing changes for personal projects, and nobody gained access they did not already have - editors and viewers are unchanged.

Closing every panel in the editor used to leave a blank screen with no visible way back - it could look like your project was gone. It never was: your work is always saved.

Now the editor keeps its header with the project menu at all times, and an empty workspace shows the AI prompt plus one-click buttons to reopen Preview, AI Chat, and the other panels. Asking the AI from there opens the chat with the answer. And if a saved project somehow ends up with no panels at all, it now opens with the standard layout instead of a blank screen.

Ask for a mobile app - or add one to a project you already have - and the agent builds it alongside your site and backend, in the same project.

• One project, three surfaces. The mobile app (React Native + Expo)

lives next to your frontend and your NestJS backend and talks to the same API and database. A mobile app that arrives with a real backend, its own data, a domain and email is the whole product, not a demo shell.

• See it in the browser first. The app renders inside the normal

Playcode preview while the agent works, so you watch screens appear without touching your phone.

• Then hold it in your hand. Scan the QR code the assistant prints in

chat and the same app opens on your real phone through Expo Go, live against your project's backend.

Store binaries and app-store submission are a separate, upcoming step - what ships today is building, previewing, and running your app on a real device.

Your project's address no longer requires a detour through a registrar.

• Search, buy, attached. Look up a name, pay with your card, and the

domain is registered to you and connected to your project - DNS records, the www variant, and the certificate handled for you.

• A guided flow that tells the truth. Every step of the purchase shows

its real state - reserved, registered, connecting, live - so you are never left staring at a spinner wondering if your money did anything.

• Your domain stays yours. It is a normal registration in your name;

nothing locks it to us.

The publish dialog could not be scrolled on a phone, which on small screens put the confirm button out of reach. The dialog body now scrolls properly, so you can publish a project from your phone end to end.

Speaking your idea on the home screen asked the wrong wallet whether you were allowed to. If you were working inside a team workspace, the microphone checked your personal plan and credits instead of the team's - so a team with an active subscription and credits to spare could still be told it needed to subscribe.

Voice now prices the same workspace everything else does: the one you have selected, which is the one a new project is created in and the one that pays for the work. Typing was never affected.

On very long builds the agent periodically folds older conversation history into a compact brief to stay fast. That fold used to be lossy: the agent could forget its instructions, lose its place, and re-read files it had just read. The whole handoff was rebuilt:

• The agent now writes its own handoff note. Instead of an outside

summary, the model working on your project writes a structured continuation brief for itself - what it was doing, what it decided, what comes next, in its own words.

• Task instructions can no longer be lost. A background task's original

instructions survive every handoff word-for-word.

• No more re-reading loops. The brief carries the exact code regions the

next step needs, pre-loaded - resuming costs zero extra reads instead of re-opening file after file.

• A stuck loop breaks itself. If context refills right after a fold, the

agent gets more room instead of folding again and again.

• A heads-up before the fold. The agent is warned as the handoff

approaches, so it finishes the step it is on and writes things down first.

Ask for something big - a storefront, an admin panel, order emails - and the agent now breaks it into parts you can follow, instead of leaving you to read a stream of technical steps.

• A plan card in the chat. The parts of your order in your own words,

each one pending, building, or done. On full-stack apps, a part only turns done after the server checks that its page actually loads - and every finished part carries an Open button with the real address, so "the admin panel is ready" is a click, not a path to guess.

• One line while it works. During a build, a quiet line above the

message box shows which part is in progress and for how long - real elapsed time, not a guess.

• The plan opens as its own page. Click it to read the full plan,

nicely rendered with diagrams, without leaving the chat. When the agent updates the plan, you see what changed - changed sections are marked, and a history lets you read any earlier version. Every plan change is a real commit in your project's git history.

• The preview stops playing dead. While the agent rebuilds your app,

the preview keeps the last good page dimmed under one line - "Rebuilding this page - it comes back by itself" - instead of flashing white. A new health check catches blank pages the old green light missed. When nothing is running, real errors still show as errors.

• The agent answers in your language. Switching the app language now

reliably carries into the agent's replies, plan titles, and progress notes.

• Bring your own instructions and skills. A project's AGENTS.md is

now read automatically by every agent working on it - same rules as other AI coding tools, whole file, no silent truncation - and projects can carry their own skills in .agents/skills/, listed and loadable like built-ins.

On long builds the agent periodically compacts older conversation history to stay fast. Until now it compacted early enough that big tasks - many files, long sessions - could lose track of files it had just read and re-read them.

• The default Context level is now High (was Medium). On models with large

context windows the agent keeps roughly 2.5x more of the conversation before compacting, so long builds stay coherent instead of looping over the same files.

• No extra confirmation for the default. The one-tap cost acknowledgment

now appears only when you raise Context above High (Extra High or Max).

• Your explicit choice is untouched. If you set Medium (or any level)

yourself, that choice stays exactly as you set it.

You can still change Context per chat in the build-quality dialog, or set your own default in Settings → You → AI defaults.

When your app emails an address that has not confirmed it wants mail yet, that message is Held - stored safely, waiting on a verification. Until today, a held message looked just like normal captured mail, and it was easy to miss that real messages were waiting on you. That is fixed, end to end:

• A new Held status. Held messages show an amber Held chip with the exact reason on the row - clearly separate from mail that is simply captured by design.
• A standing notice in the editor. While anything is held, the panel shows "N messages waiting on you" with a direct door to Settings > Email.
• An email to you. The first time a message is held for an address, the project owner gets an email naming the address - so a contact form filling up overnight can no longer wait silently. Repeats fold into at most one summary a day.
• Deliver held mail in one click. Once the address verifies, Settings shows how many messages were held and a Deliver held button. Nothing ever sends without you.

Two more changes in the same area:

• The Mailbox panel is now called Emails - a clearer name for what it is: every email your app sent, with its delivery status. Playcode Email sends mail; it does not receive it.
• A new help page, What emails can your app send, explains the transactional-only sending policy in plain language.

When the assistant starts a long job - a build, an app-store submission, a big install - the chat now stays yours.

• You can always talk. A message you send while a job runs reaches the

assistant immediately, mid-work, instead of waiting in silence until the job ends.

• Stop means stop. Pressing Stop lands in about two seconds, even during a

long wait - and the assistant stays quiet afterwards until you speak.

• Finished jobs report back on their own. When a background job completes,

a small note appears in the chat and the assistant picks the result up and continues - even if it had already handed the chat back to you. A running job can no longer be forgotten.

• One question, one answer. A message sent at just the wrong moment no

longer risks two competing replies.

When something has to reach your phone - opening your app in Expo Go, checking a preview on a real device - the assistant now draws a QR code in the chat instead of asking you to type a long address across two screens.

• The address stays printed under the code as selectable text, so on a desktop

you can still copy it in one click.

• The code carries the Playcode mark in the middle, and it is rendered to be

scannable rather than decorative: it sizes itself to the address it holds, and in a column too narrow to scan from it steps aside and leaves you the text.

• It works the same in dark and light themes.

Only for addresses that genuinely have to cross to another device. A link you can simply tap stays a link.

If you were signed in but arrived on a device or browser Playcode had never loaded before - a new laptop, a cleared browser, a private window - opening a template link directly (/javascript-template and friends) could leave you staring at an empty editor: no files, no preview, no menu. Reloading did not help.

The editor was starting before your account finished arriving from the server, and then quietly gave up. Two things changed:

• The editor no longer waits on that data to build your project, so a first

visit on a new device opens the same way a returning one does.

• If an open genuinely fails, you now get a clear message with a Retry button

instead of a blank page.

When you upgrade to Max, every project you already have now moves to the larger Max machines automatically - 8 CPU · 16 GB in the editor and bigger production machines with a 200 GB disk - not only the projects you create after upgrading.

• Projects that are asleep pick up the new hardware silently and wake up on

it the next time you open them.

• An editor you are actively using switches on its next idle pause, usually

within minutes.

• Published production apps pause briefly - usually seconds, at most a couple

of minutes - while they move to a bigger machine.

And if you ever move back to Pro: CPU and RAM return to Pro sizes, but any disk space you grew into stays yours. We never shrink your disk.

We reshaped the paid plans around one idea: a dollar buys the same 4 AI credits everywhere, and the plans differ in how much you get - not in what a credit costs.

• Pro is now one simple plan. $25/month, 100 credits, 3 full-stack cloud

apps, custom domains. No size picker - a heavy month is covered by top-ups.

• Meet Max (formerly Business). For people who need more of everything:

5 cloud apps, larger VMs (8 CPU · 16 GB editor), 1 TB/month bandwidth - and now with double the credits at the same prices. $100/month includes 400 credits, $200 includes 800, $400 includes 1,600, and new sizes reach $2,500/month with 10,000 credits.

• Top-ups are one rate for everyone. $1 = 4 credits on every plan, with

the same volume bonuses (+5% at $50, +10% at $100, +20% at $200).

Existing subscribers keep their current plan and price; Business subscribers get the doubled credits automatically from their next renewal.

Connecting a custom domain to a Cloud app now explains itself better - especially for root domains.

• Squarespace DNS is now recognized. If your domain runs on Squarespace

nameservers, the wizard shows the exact ALIAS record their panel expects.

• The wizard names your DNS provider. When a root domain can't point at

Playcode, the wizard now tells you which DNS service it detected and which nameservers prove it - no more guessing whether the check is right. This especially helps domains bought at Google Domains and moved to Squarespace, which often still run on Google's nameservers behind the scenes.

• Unknown providers are never a dead end. If we can't identify your DNS

host, the wizard now shows the ALIAS record to try instead of blocking you - and still recommends the www address that works everywhere.

Browser projects now have Checkpoints - restore points you can save at any moment and return to instantly.

• Save a restore point. Open the project menu and choose **Checkpoint

history, then New checkpoint**. It captures all your files exactly as they are - instantly, without using any credits.

• Name your checkpoints. Rename any checkpoint so "Before the redesign"

is easier to find than a date.

• Restore instantly - and undo it. Restoring puts every file back the way

it was. Playcode saves a safety checkpoint of your current state first, so a restore is never a one-way door. The page does not reload, and your chats stay exactly where they are.

• Publish keeps you covered. Every publish automatically saves a

checkpoint of what went live, so you can always get back to a published state.

You can keep up to 20 checkpoints per project. At the limit, Playcode asks before making room by deleting the oldest one.

Want to explore two directions side by side? Use Duplicate project - the copy gets its own files, chats, and published site.