Bolt.new Pricing in 2026: What Tokens Actually Cost

Bolt.new pricing at a glance

These are the published numbers on Bolt's own pricing page, read live on 18 September 2026. Bolt is a StackBlitz product and the page is the only source that reflects current prices, so it wins over any third-party summary.

The Pro 50 and Pro 100 figures come from Bolt's billing documentation, which walks through an upgrade example using exactly those plan names, prices and token allotments. The pricing page itself shows one Pro entry point and lets you scale up from there.

Annual billing is the one place the sources disagree. The pricing page advertises savings of up to 28% with yearly billing, while the billing help article describes the annual toggle as 10% off, and separately quotes a Pro 10M annual plan at $216 per year. $216 against $25 per month is a 28% saving, so the pricing page number is the one that matches the worked example. Check the number in your own checkout before committing to a year.

Reading further into the same billing article narrows it. Its worked examples quote a Pro 10M annual plan at $216 a year and a Pro 26M annual plan at $540. Against $25 and $50 billed monthly, that is 28% off the first and 10% off the second, so the discount varies by tier and both published figures can be true at once. The entry tier is where the saving is largest.

What a token actually costs on each Bolt.new tier

Dividing the price by the allowance is the only way to compare these plans honestly, and it produces a result the plan names hide.

The annual and monthly figures come from the worked upgrade examples in Bolt's billing documentation, which quotes a Pro 10M annual plan at $216 a year and a Pro 26M annual plan at $540.

Three things worth taking from that table.

Scaling up is a real discount, unlike most credit systems. Moving from Pro to Pro 100 cuts the cost per million tokens by 27%. If you know you are going to spend it, the higher tier is genuinely cheaper per unit of work rather than just larger.

Teams costs 20% more per token than Pro, on the 10M starting allowance per member shown in the plan table above. $30 a member against $25 for Pro. The difference buys centralised billing, admin controls, user provisioning, private npm registries and design-system prompts. It does not buy capacity, and Bolt's Teams billing documentation confirms tokens are not shared between members, so a team of five is five separate allowances rather than one pool of 50M. Admins can see each member's allocation and spend in the Consumption by members table.

The Pro 200 tier exists. Bolt's billing article walks through a downgrade from a Pro 200 monthly plan, so the ladder continues past Pro 100, though that article does not state its token allowance. The selector in your own account is the place to read it.

How token billing actually works

Every Bolt.new plan is a subscription that bundles a monthly token allowance. Tokens are the meter. You are not paying per project, per seat capability or per deploy, you are paying for how much work the model does on your behalf.

The part that surprises people is where the tokens go. Bolt's own FAQ is blunt about it: most token usage comes from syncing your project's file system to the AI, so the larger the project, the more tokens each message costs. That means the same prompt gets more expensive over the life of a build, not because the prompt changed but because the codebase around it grew.

Two structural consequences follow. First, cost rises with project size even when your habits stay identical. Second, the expensive part of a session is often not the code you wanted written, it is the context Bolt had to load to write it.

What burns tokens fastest

Bolt's token efficiency guide names the specific levers, and they map cleanly onto where budgets actually go:

• Project size. Bolt processes fewer tokens when it has less context to process. A 40-file project costs more per message than a 10-file one.
• Agent choice. The Max agent works through a fuller scope and uses more tokens than the Standard agent. Using Max for routine edits spends more than the task needs.
• Build mode versus Plan mode. In Build mode every prompt updates code and uses tokens. Plan mode is for talking through decisions and uses fewer, because it changes nothing.
• Unfocused prompts. Pointing the agent at specific files or functions costs less than letting it consider the entire codebase.

The pattern most buyers report is the debugging loop: a fix that does not land, followed by another attempt over the same enlarged context. One builder's March 2025 write-up puts a number on both ends of the range, describing roughly 100K tokens per prompt as acceptable on a working project and around 400K per prompt once the project got large. That is a 4x swing driven by context growth rather than by anything the user chose.

How to make your Bolt.new tokens last

Bolt publishes more practical guidance here than almost any vendor in this category, and most of it is free to act on. Everything below is from Bolt's own documentation.

Do the free things in the interface instead of asking for them. Clicking an on-screen button or action costs nothing, prompting costs tokens. Use the Publish button rather than asking Bolt to publish. Use Version history to roll back rather than asking Bolt to undo. Edit existing code directly in Code view, which does not consume tokens at all and still works after you have hit the free plan's daily cap.

Stop clicking Attempt fix. Bolt's guidance is explicit: each attempt costs tokens, and clicking it repeatedly in hope is the pattern that empties an allowance. If two attempts have not landed it, switch to Plan mode and work out the cause before spending again, or read the error yourself.

When you are stuck in a loop, ask for logging rather than a fix. Bolt recommends prompting it to add detailed error logging around the problem area. The next failure then carries real information, so the fix attempt after that has something to work from. One paid detour beats five paid guesses.

Clear the context when you finish a piece of work. Typing `/clear` resets Bolt's understanding of the conversation, and Bolt's context guide says requests then use fewer tokens. The codebase and files are still there, Bolt simply reads them fresh. The natural moment is when you finish a feature, not when you notice responses getting slow.

Move permanent facts out of the conversation. Project knowledge and an `agents.md` file in the project are both persistent, so Bolt can reach them without you restating them in every prompt, and neither is wiped when you clear context. Bolt finds `agents.md` automatically once it is in the project.

Invert your prompt length as the project grows. A long, detailed first prompt is efficient on an empty project. The same long prompt on a mature project pushes an already full context window. Bolt's advice is to break large requests into smaller focused ones as the codebase grows.

Delete the files Bolt wrote for itself. As it builds, Bolt adds Markdown files for record keeping, then reads them into context on every subsequent prompt. Bolt suggests asking it to keep only the `.md` files that carry necessary project context. For unused files generally, it recommends either prompting for a cleanup or running `npx knip --production --fix --allow-remove-files` in the Code view terminal, which uses no tokens because it is a terminal command rather than a prompt.

Split files over about 700 lines. Bolt's documentation gives the exact prompt for listing every file over 700 lines, then refactoring them one at a time with the original kept as a router. Refactoring costs tokens up front and reduces them on every message afterwards, because Bolt reads less code per change.

Match the agent to the task. Bolt has two agents, Standard and Max. Standard is described as fast and token efficient and is the better default for well-defined work. Max reasons more and costs more, and Bolt says that on tasks where Standard already performs well, Max may not show a noticeable difference. Only the Standard agent is available on the free plan, and Bolt remembers your choice per project.

Turn connectors off when you are not using them. Connectors add external context, and that context is tokens on every message. Bolt recommends switching them on per project only when needed.

Consider Bolt Lite if tokens are the whole constraint. Bolt Forge runs open-source models with their own usage allowance, and per Bolt's Forge documentation it is included at no extra cost on every individual Pro plan during the preview, or available through Lite, a standalone plan at $9 a month, by waitlist. Two caveats read on 18 September 2026: Forge is not available on Teams or Enterprise plans, and that page says Forge access on Pro plans ends on October 14, after which Lite subscribers keep the $9 plan and Pro users continue on Standard and Max. Check the current state of the preview before planning around it.

Do Bolt.new tokens roll over?

Yes on paid plans, no on free. This is the detail worth getting right, because it changes how you size a plan.

Per Bolt's token documentation, unused tokens from a paid monthly allocation roll over and stay valid for two months from the start of the billing cycle that issued them. Consumption is first in, first out, so the oldest rollover bucket is spent before the current month's allocation. Free plan tokens do not roll over at all, and the 300K daily cap does not accumulate either. Not using today's 300K does not give you 600K tomorrow.

Three conditions matter. Rollover requires an active paid subscription, so cancelling forfeits access to banked tokens when the cycle ends, though unexpired tokens come back if you resubscribe in time. Upgrading never destroys existing tokens. And separately purchased reload tokens do not expire, but reloads are only offered on the highest individual monthly Pro plan or on any annual Pro plan.

What an active builder actually pays per month

Here is the arithmetic, with the assumptions stated so you can swap in your own.

Assume a solo builder working on one real project most weekdays, roughly 60 substantive prompts a month. On a small project at around 100K tokens per prompt, that is 6M tokens, comfortably inside Pro's 10M at $25. On a grown project at around 400K tokens per prompt, the same 60 prompts is 24M tokens, which needs Pro 50 at $50 and leaves almost nothing spare.

Add a normal share of failed fixes and the honest planning number for someone actively building, not evaluating, is $50 to $100 per month, meaning Pro 50 or Pro 100. The $25 entry price is real, but it describes a smaller project than most people have by month three. Rollover softens a quiet month, it does not rescue a heavy one, since banked tokens expire in two months.

Four worked scenarios, with the assumptions stated

This is arithmetic, not a survey. The two per-prompt figures come from the builder write-up cited above, roughly 100K tokens per prompt on a healthy project and roughly 400K once it has grown, and everything else is stated so you can substitute your own numbers.

The free plan's constraint is the daily cap, not the monthly total. 300K tokens is one substantive message on a project of any size, and unused daily tokens do not carry into tomorrow. Bolt's own documentation makes the point directly: not using today's 300K does not give you 600K tomorrow.

The gap between the second and third rows is not a change in habits. It is the same person, working the same way, six months later on a larger codebase. That is the single most important thing to understand about Bolt.new pricing, and it is why sizing a plan from your first month is usually wrong.

A team of three on Pro-level Teams seats is $90 a month for 10M tokens each, and those allowances do not pool. If one person is doing most of the building, the money is in the wrong place.

Bolt.new pricing compared to Lovable, Replit, v0 and Cursor

Every tool here meters something, and no two meter the same thing, which is why entry prices are close to meaningless side by side. These figures were read live on 18 September 2026 from each vendor's own pricing or billing pages.

What that table means in practice.

Bolt is the only one that charges you for the size of your codebase. Lovable bills per unit of work completed and publishes example costs per prompt. Replit bills per completed checkpoint under what it calls effort-based pricing. Bolt bills for tokens processed, and most of those tokens are your own project files being read back to the model. Identical habits cost more in month six than in month one.

v0 publishes the clearest rate card in the category. Its pricing page lists input and output prices per million tokens for each model, from $0.20 and $1.20 on v0 Mini to $10 and $50 on v0 Max Fast, and credits are denominated in dollars. If forecastability is what you want from a builder, that is where it currently lives.

Cursor tells you the real number instead of the entry price. Its pricing documentation says daily agent users typically land at $60 to $100 a month of total usage and power users often exceed $200, against a $20 Pro plan. Pro Plus is $60, Ultra is $200, and Teams seats are $40 or $120 per user. It is a code editor rather than a browser app builder, so treat it as a budget comparison.

Hosting is bundled on Bolt and metered on Lovable and Replit. Bolt includes Pro hosting with every paid subscription, up to published bandwidth and request limits, with optional pay as you go beyond them. On Lovable and Replit, a live app draws down the same balance you build with. That is the difference between a bill that grows while you build and a bill that grows when you succeed.

Where the published numbers disagree

Third-party pricing pages for Bolt.new are unusually inconsistent. Live checks against Bolt's own page turn up summaries still quoting Pro at $20, and several assert that tokens do not roll over, which contradicts both the pricing page FAQ and the token help article. Bolt's FAQ dates the rollover change to 1 July 2025, which is probably when those pages were written and never revised.

The practical rule: for anything token related, read Bolt's pricing page and help center, and treat everything else as a starting point for questions. That is also the honest reason this post cites Bolt's own documentation for every number above.

What changed in Bolt.new's pricing, and why it matters

Those stale pages are stale about two specific things, and it is worth naming them, because the changes cut in opposite directions.

Rollover arrived on 1 July 2025. Before that, unused tokens simply expired. Bolt's pricing page FAQ dates the change precisely, and the token documentation fills in the mechanics: two months of validity, first in first out, and an active paid subscription required to touch banked tokens. Any page that tells you Bolt tokens never roll over was written before that date or copied from something that was.

The entry price moved up to $25. One concrete example of what is still circulating: a 2025 review of Bolt describes paid plans beginning at $20 a month for 10 million tokens, lists Pro 50 at $50 for 26M and Pro 100 at $100 for 55M, and states that subscription tokens do not roll over. Those two tier figures match Bolt's billing documentation exactly. The entry price and the rollover claim contradict Bolt's live pricing page. That mix is the trap: a page can be right about the shape and wrong about the two numbers that decide your bill.

So the price went up and the tokens became worth more, because they now survive an extra month. A quiet month followed by a heavy one is genuinely cheaper than it used to be. What did not change is the thing that actually drives the bill, which is that project size sets the cost of a message.

There is also a live promotion to know about. Bolt Forge, which runs open-source models with a separate allowance, is included with individual Pro plans during its preview, and Bolt Lite exists at $9 a month as a standalone way to get it. Bolt's Forge page, read on 18 September 2026, says Forge access on Pro plans ends on October 14. Anything you read about Bolt costing $9 is describing that plan, not the app builder most people mean.

I built a prototype. What does it cost to take it to production?

Bolt handles this differently from the credit-metered tools, and mostly in your favour.

Hosting is bundled. Per Bolt's hosting plans documentation, the free level covers up to 10 GB of bandwidth and 333,333 requests a month, and Pro hosting, included with every paid subscription, covers up to 30 GB and 1 million requests. Bolt puts rough numbers on that: the free level typically supports around 10,000 to 15,000 monthly visits, and Pro can often support 100,000 or more, varying with page weight.

Three details decide whether that is enough.

The limits are per account, not per project. Publish five sites and they share one bandwidth and request allowance.

A free site goes offline when it hits the cap and stays offline until usage resets at the start of the next billing cycle. On Pro you can turn on pay as you go, which buys extra bandwidth and requests automatically up to a spending cap you set. Once that cap is reached the site pauses too, so the cap is the number to set before a launch rather than during one.

The Made in Bolt badge is on free projects and clears on the next publish after you upgrade.

Everything else about going to production is a normal engineering decision rather than a billing one. Bolt connects to GitHub for version control, publishes to a `bolt.host` address or a custom domain you buy or connect inside the product, and can deploy through Netlify instead if you would rather own the hosting relationship. The database and server functions live in Bolt Cloud.

The honest caveat, and the one most reviewers land on: what Bolt produces is a fast, working prototype, and the code usually needs engineering review before it carries a real user base. Budget for that as time rather than as tokens.

Who Bolt.new is genuinely wrong for

Bolt is the fastest way I have seen to get from a prompt to a deployed, working web app. These are the cases where that is the wrong trade.

Anyone maintaining a large codebase for a long time. The cost model works against you by design, because every message pays to sync a project that only grows. Bolt's own efficiency guidance is largely a list of ways to make your project smaller. If the project is supposed to get bigger, you are fighting the meter permanently.

Teams that need pooled budgets. Teams tokens are allocated per paid member and are not shared, so an unevenly loaded team overpays. There is no way to give the person doing the building the allowance of the four people who are not.

Anyone who cannot read the code when it breaks. The recurring failure in reviews is the model breaking a working feature while fixing another, with each recovery attempt costing tokens. A review published August 26, 2026 describes exactly that past a dozen or so components, and quotes a Trustpilot reviewer saying most of their tokens went on fixing repeated issues. That publisher sells a competing builder, so weigh it accordingly, though the pattern matches what Bolt's own guidance implies when it tells you to stop clicking Attempt fix. A developer reads the diff and steps in. A non-technical builder has only the option that costs money.

Budget owners who need a fixed monthly number. The same person, doing the same work, can spend 4x more in month six than in month one. You can plan a floor. You cannot plan a ceiling without setting one deliberately and living inside it.

Projects where the live app must never pause. Free hosting stops serving at the cap, and Pro pay as you go stops at your spending cap. Both are correct behaviour for a budget control and both are outages if nobody set the number with a launch in mind.

Where this sits next to your AI tooling stack

Bolt.new is an app builder. You prompt it, it writes and runs a project in the browser. Its cost model is a direct function of how much code the model reads and writes for you, which is why token burn is the whole conversation.

That is a different line item from connecting AI tools to the systems your company already runs on. If you are also comparing what it costs to give Claude Code, Cursor, ChatGPT or Windsurf real context across your internal tools, that is a separate budget with a separate shape. Coworker MCP connects AI tools to 50+ apps through one governed connection at published pricing, Pro at $29.99 and Max at $149.99 per user per month with custom Enterprise, rather than a usage meter that moves with project size. It solves an adjacent problem to Bolt: Bolt generates an app, Coworker MCP gives your AI tools context and access across the systems you already have.

If that comparison is the one you are actually running, what MCP is is the right starting point, and how to choose MCP servers covers the evaluation criteria. For per-seat AI tooling in the same budget line, see our pricing comparison and the breakdown of Windsurf pricing.

Book a demo to see how Coworker MCP prices against your current AI tooling spend.

Frequently asked questions

How much does Bolt.new cost?

Bolt.new has a $0 free plan with 1M tokens per month, Pro from $25 per month with at least 10M tokens, Teams at $30 per member per month, and custom Enterprise pricing. Higher Pro tiers run $50 for 26M tokens and $100 for 55M tokens.

Do Bolt.new tokens roll over?

On paid plans, yes. Unused tokens roll over and remain valid for two months from the start of the cycle that issued them, and the oldest tokens are spent first. Free plan tokens do not roll over, and the 300K daily cap resets rather than accumulating.

What uses the most tokens in Bolt.new?

Syncing your project files to the model, according to Bolt's own FAQ. That makes project size the dominant cost driver. Using the Max agent for simple tasks, prompting in Build mode when Plan mode would do, and broad prompts across the whole codebase all add to it.

Is the Bolt.new free plan enough to build a real app?

For a single page or a proof of concept, often yes. The 300K daily cap is the real constraint, because a growing project can spend a meaningful share of that on one substantive message, and free plan tokens do not carry into the next day.

What happens to my tokens if I cancel Bolt.new?

Access to allocated and rolled over tokens ends when the billing cycle ends. If you resubscribe before those tokens reach their two month expiry, unexpired balances become available again alongside your new allocation.

Why do other sites list Bolt.new Pro at $20?

Because they are out of date. Bolt's live pricing page lists Pro at $25 per month billed monthly. Several third-party summaries also still state that tokens never roll over, which Bolt changed on 1 July 2025.

How can I reduce token usage in Bolt.new?

Bolt's own list, in rough order of impact: use buttons rather than prompts for anything the interface can do, including Publish and Version history, since those cost nothing. Edit existing code in Code view, which uses no tokens. Clear context with `/clear` when you finish a feature. Put permanent facts in project knowledge or an `agents.md` file instead of restating them. Keep prompts short and file-specific as the project grows. Delete unused files and split files over about 700 lines. Use the Standard agent unless the task genuinely needs Max. Stop clicking Attempt fix after the second try.

Does Bolt.new hosting cost extra?

No. Pro hosting is included with every paid subscription, covering up to 30 GB of bandwidth and 1 million requests a month across your whole account. The free level is 10 GB and 333,333 requests, and a free site stops serving when it hits those limits until the next cycle. Pro users can enable pay as you go for traffic beyond the allowance, up to a spending cap they set.

What is the difference between the Standard and Max agents?

Standard is faster and more token efficient and is the better default for well-defined work like UI changes and small to medium apps. Max reasons more thoroughly, which helps on large codebases, interconnected features, refactoring and open-ended problems, and costs more tokens. Bolt notes that on tasks where Standard already does well, Max may not produce a noticeably different result. Only Standard is available on the free plan.

Is Bolt Teams worth it over individual Pro plans?

Only for the administration. Teams is $30 per member per month against $25 for Pro, for the same 10M starting allowance, so you pay about 20% more per token. What you get is centralised billing, access management, user provisioning, private npm registries and design-system prompts. Tokens are not shared between members, so Teams does not help an unevenly loaded team.

What is Bolt Lite and is it cheaper than Pro?

Bolt Lite is a standalone $9 a month plan that gives access to Bolt Forge, which runs open-source models with their own usage allowance, offered by waitlist. Forge is also included with individual Pro plans during the preview. Bolt's documentation, read on 18 September 2026, says Forge access on Pro plans ends on October 14 while existing Lite subscribers keep the $9 price. It is a cheaper way to build with different models, not a cheaper version of the Pro plan.

Related reading

• What Is MCP?
• How to Choose MCP Servers for Your Team
• Enterprise AI Pricing Compared
• Windsurf Pricing
• Claude Code Pricing
• How Much Should Enterprise AI Cost Per User?