Pricing & tokens

Instead of mystery “credits,” VibeX shows the actual tokens used behind each request and what they cost, so you always know what a build costs.

• Start free with a 2M metered-token allowance — a one-time lifetime amount to try VibeX.
• Pro adds 30M metered tokens a month; Pro Max adds 60M a month.
• Usage is metered per request and shown live — you always see the real cost.

The cost the meter shows is the estimated AI-provider cost for each request. It's included in your plan's token allowance — you're billed your subscription, never per request.

Allowances are hard caps

Your token allowance is a hard cap, not a soft target. When you reach it, new builds pause — they're blocked before anything runs, so a paused build never costs you tokens. The build already in progress when you cross the limit completes normally (its size is bounded by the per-build cap), and everything you've made stays available: your projects, checkpoints, exports, and published apps are unaffected.

• Pro and Pro Max: building resumes automatically when your monthly allowance renews on the 1st (UTC), or immediately if you upgrade.
• Free: the allowance is a one-time lifetime amount — upgrade to keep building.

How tokens are weighted

Different token categories have different AI-provider costs. Fresh input is $2 per million tokens, cached input is far cheaper — $0.20 per million to re-read context the provider already has (and $2.50 per million to write it to the cache) — and output is $10 per million. The live meter shows the real per-category breakdown for every request, so nothing is averaged away.

Your plan's allowance counts metered tokens: each category is weighted in proportion to that real cost, from the same rate table the meter uses. Today one fresh input token counts as ~2.4 metered tokens, a cache write as ~3, an output token as ~12, and a cached read as just ~0.24 — re-reading context the provider already has barely touches your allowance.

Because most of a typical build's input is cached context, real builds usually consume fewer metered tokens than their raw count — the cache working in your favor, live on the meter. The weighting exists for the other direction: it makes sure no unusual, output-heavy usage pattern can quietly cost more than your plan includes.

The Successful Build Guarantee

If VibeX cannot produce a compiling, previewable result for your request because of an engine or platform failure, that attempt does not count against your included usage. You'll see what failed and whether completed work was preserved.

What counts as an engine or platform failure — and what doesn't:

• Covered (never debited): our engine or platform errors and can't deliver a result — the AI provider fails mid-build, our pipeline errors, or anything on our side stops the build from completing. Usage is only ever debited when a result is delivered, so these attempts never touch your meter.
• Not covered (normal usage): a build that completes but isn't what you wanted — refining it by chatting is normal, metered usage (checkpoints make iteration safe).
• Not covered: requests VibeX declines for safety reasons, or requests outside what VibeX supports today (see What VibeX builds) — the work the engine did on them is metered.
• Third-party outages: if an outside outage stops your build from completing, you aren't debited (same principle — no delivered result, no debit). The guarantee's promise is about our engine and platform; we can't guarantee third-party uptime itself.

Published-app limits

Each plan can keep a set number of apps published at once: Free 3, Pro 25, and Pro Max 50. When you reach your limit, unpublish one to publish another, or upgrade for more slots — your projects are never deleted, and publishing a new version of an existing app doesn't use another slot.

See the plans on the pricing page. Every plan — including free — lets you export your code to GitHub.