FitbuildBuild apps with AI, no coding experience needed
Menu

Best free AI coding tools

Free is a different product on every one of these tools. Some free plans can carry a small web app. Some are a short look at an editor. One famous name is not a coding tool until you pay. The numbers below are the free plans, checked 2026-10-19. Nothing here invents a quota the vendor did not publish.

When the meter hits zero, go to when you hit your limit before you open a second account. Waiting is right when the reset is soon. It is wrong when you will hit the same cap again next week.

The call

  • A browser app, and you want the free plan most likely to survive past a demo: Bolt.new. Lovable if you accept a hard monthly cap and no code download. v0 if you only need the interface. Replit if you want the IDE and a database and you can live with an unpublished daily credit number.
  • An editor, and you do not want a subscription yet: Google Antigravity. Cursor Free, Windsurf, and Kiro Free (50 credits a month) are shorter trials. Amp Hobby is $0 to Amp if you bring a model subscription and your own machine. Copilot's free tier is real, and the exact allowance is marked verify because GitHub's pages disagree.
  • Claude Code: do not start on free. Free Claude does not include Claude Code. You need Pro or Max before this tool exists for you. The paid floor is on the tool page. Jules is the free plan with a published daily task count, and it returns a pull request rather than a hosted app.

Enough to start something small

Free-tier weight 3/5 or higher, and the free plan actually includes the tool. Weight is an editorial score, the same kind the quiz uses. It is not a benchmark.

Bolt.new

Free-tier weight 4/5 · checked 2026-10-19

A browser builder with a generous free token pool and a $25 Pro plan.

Free · Free Verified

1 million tokens per month and a 300,000 token daily cap. Public and private projects, Bolt branding, and a 10MB upload limit. Direct edits in Code view do not spend tokens.

300K tokens per day and 1M tokens per month. Bolt branding, 10MB uploads, hosting up to 333k web requests.

Daily leftover tokens do not carry into the next day.

Enough for. A first web app or landing page for someone who wants a month of building, including private projects, rather than a short demo.

When free dies. When the daily cap or the monthly pool on the card is gone. Leftover daily tokens do not carry forward. Direct edits in Code view do not spend tokens, which is the move before you pay. A large project spends more per message. Reset: Free tokens reset on the 1st and do not roll over. The daily 300K cap resets every day and also does not roll. Paid tokens reset on the renewal date and roll for one extra month while the subscription stays active.

What to do when Bolt.new runs out

Google Antigravity

Free-tier weight 4/5 · checked 2026-10-19

A free agent IDE, with higher quotas on Google AI Pro and Ultra.

Individual · Free Verified

Free. Unlimited Tab completions and Command requests, Gemini models, and a weekly agent quota. The quota is not a published prompt count. It refreshes weekly, not every five hours.

Unlimited Tab and Command requests. Weekly agent quota. Gemini models, including Gemini 3.8 Flash, which the plans doc names as a core agent model. No published prompt count.

The Antigravity pricing page lists this at $0/mo.

Vendor does not publish a fixed count.

Enough for. Trying a real editor agent without a subscription, if you can follow the changes and you can wait out a weekly quota.

When free dies. When the weekly agent quota is gone. Tab and Command stay unlimited on this plan. There is no credit overage, and the vendor does not publish the agent quota as a count. Reset: Free quota refreshes weekly. Google AI Pro refreshes every five hours until a weekly limit. Ultra refreshes every five hours and has the highest weekly limits. Prompt counts are not published because heavier tasks spend more quota.

What to do when Google Antigravity runs out

Jules

Free-tier weight 4/5 · checked 2026-10-19

An async agent that takes a GitHub task and opens a pull request.

Jules · Free Verified

15 tasks per rolling 24 hours, 3 concurrent, on Gemini 2.5 Pro. Jules describes this plan as evaluating the agent on real work. It does not host the project.

15 tasks per rolling 24 hours, 3 concurrent tasks, Gemini 2.5 Pro. Jules' own label for this row is evaluating the agent on real work.

Task counts are on jules.google and the usage-limits doc. Neither page prints a dollar price because this row is free.

Enough for. Real tasks on a GitHub repo you already have: 15 a day, 3 at once, and a pull request to review. It does not host the app.

When free dies. At task 16 in the rolling 24 hours, or when a fourth task would run at the same time. The window is not midnight. Reset: Task limits are a rolling 24 hours, not midnight. Concurrent-task caps apply at the same time. Family-plan members each have their own limits. They are not pooled.

What to do when Jules runs out

Amp

Free-tier weight 3/5 · checked 2026-10-19

A coding agent you can run free on your own machine, or in Amp's cloud.

Hobby · Free Verified

Free to Amp if you bring your own model subscription or key and run on your own machine. Cloud orbs are pay-as-you-go. Hobby includes no orb minutes and no Amp credits.

All product features. No monthly fee. Pay-as-you-go orbs, or your own runners at no Amp charge. Linked ChatGPT, Claude, and other subscriptions, and bring-your-own-key, with no Amp token fees or limits. No included orb minutes and no included Amp credits.

The Hobby card is on ampcode.com/pricing. Standard orb rates, if you do use cloud machines, are on the orb sizes page and start at $0.08 an hour for a1.tiny.

Enough for. Real agent work at $0 to Amp, if you already have a ChatGPT or Claude subscription and you can run the agent on your own computer.

When free dies. As soon as you need Amp to pay for the model, or you need Amp's cloud machines. Hobby includes neither orb minutes nor Amp credits. Orbs on this plan are pay-as-you-go. Reset: Included Megawatt and Gigawatt usage resets at the end of the monthly billing period and does not roll over. Purchased credits expire 12 months after you buy them. A paused orb costs nothing. Amp pauses an idle orb 5 minutes after the agent stops, or 20 minutes after your last interaction, whichever is later.

What to do when Amp runs out

GitHub Copilot

Free-tier weight 3/5 · checked 2026-10-19

An IDE assistant at $10 a month, if you already live in GitHub.

Free · Free Verify

Free is real, with a tight cap. The plans FAQ still says 2,000 completions and 50 chats. A later billing post describes limited completions plus limited chat and agent usage. Confirm the live counter.

A limited free tier. The plans FAQ still lists 2,000 completions and 50 chat requests. The billing docs say 2,000 completions a month plus an AI credit allowance whose size is not stated, with auto model selection only.

The free price is real. On 2026-10-19 the chat allowance and the credit allowance still disagree across GitHub's own pages, so this row stays verify.

Enough for. A tight trial inside an IDE you already use. It is not a browser builder, and it is not a month of agent work.

When free dies. When the free allowance hits zero. GitHub's own pages describe that allowance two ways, so the row is marked verify. Read the limit line and the live counter before you plan a sprint around it. Reset: Monthly. Included credits do not roll over. Base credits are used before the flex allotment.

What to do when GitHub Copilot runs out

Lovable

Free-tier weight 3/5 · checked 2026-10-19

A browser builder for web apps, with a hard monthly cap on the free plan.

Free · Free Verified

5 build credits a day, capped at 30 per calendar month, plus 20 Cloud credits, 4 AI credits, and 1 connector credit. You cannot edit or download the code, or attach a custom domain, until Pro.

5 build credits per day, maximum 30 per calendar month. 20 Cloud credits, 4 AI credits, and 1 connector credit per month. No code edit, code download, or custom domain.

Daily credits resume on the 1st at 00:00 UTC after the monthly cap.

Enough for. A small hosted web app or site, if you can work inside a hard monthly build-credit cap and you do not need the code yet.

When free dies. When the daily build credits stop for the month. They come back on the 1st. Code edit, download, and a custom domain are not on this plan, so free dies earlier if you need the repo. Reset: Daily build credits reset at 00:00 UTC and do not roll over. On Free they stop after 30 in a calendar month until the 1st. Paid monthly credits last two months from issue.

What to do when Lovable runs out

Replit

Free-tier weight 3/5 · checked 2026-10-19

A browser IDE and agent, closest thing here to an all-in-one starter.

Starter · Free Verified

Daily Agent credits and Free Mode with a daily cap, a built-in database, and one published project. The pricing page does not state one universal credit number.

Daily Agent allowance and Free Mode with a daily cap. One published app, listed as 30-day on the Core comparison table. Built-in database. Exact daily credit count is not a single published number.

The 30-day publish window is on the Core plan comparison table. The daily credit count is still not one published number.

Vendor does not publish a fixed count.

Enough for. Learning in a browser IDE with a built-in database and one published app, if you can look at files.

When free dies. When the daily allowance is gone, or when you need a second published app. The vendor does not publish one universal credit number. Confirm the live publish rule before you depend on the free app staying up. Reset: Starter keeps a daily cap. Core and Pro Free Mode allowances reset every five hours and also have weekly limits, per Replit's plan docs. Core credits reset each billing cycle and do not roll over. Pro credits roll for two months.

What to do when Replit runs out

v0

Free-tier weight 3/5 · checked 2026-10-19

A browser UI builder for React interfaces and landing pages.

Free · Free Verified

7 messages a day, plus a credit offer that depends on the account. New signups under the one-time offer get $5 once. Existing free users may still have $5 a month. Check billing settings.

7 messages a day. Credit grant is the current offer: $5 once for new accounts on the one-time offer, or $5 a month for existing free users. Confirm which offer your account has.

The message cap is documented. The credit offer is account-specific, which is called out in the limits rather than guessed.

Enough for. A React landing page or interface, a few generations at a time, for someone who wants to see the UI before thinking about a backend.

When free dies. The same day, if you regenerate the whole page. The message cap is documented. The credit grant depends on the account, so the card tells you to confirm which offer you have. Reset: Free is capped at 7 messages a day. Paid plans are credit-based. Unused monthly credits roll one cycle and expire after 65 days. Purchased credits expire after one year and need an active paid plan to spend.

What to do when v0 runs out

A trial, not a month of work

The free plan is real and short. Use it to see the product. Do not plan a project around an unpublished cap.

Cursor

Free-tier weight 2/5 · checked 2026-10-19

An editor with agents, built for people who can read a diff.

Free · Free Verified

Free includes limited Agent and Chat, plus Tab completions on Cursor's own models. Cursor does not publish a request count.

Limited Agent and Chat requests, and Tab completions on Cursor models (Grok and Composer). No published request count. Third-party models are not available.

The current free plan on cursor.com/pricing. Legacy Hobby is the same kind of limited trial for older accounts.

Vendor does not publish a fixed count.

Enough for. Seeing limited Agent, Chat, and Tab, long enough to learn whether you can read a diff.

When free dies. When the unpublished cap stops the work. Cursor does not publish a request count for Free. Reset: On The X Subscription the included pool refreshes weekly, and unused included usage does not roll over. Legacy Cursor plans reset monthly with the billing cycle.

What to do when Cursor runs out

Kiro

Free-tier weight 2/5 · checked 2026-10-19

An agentic IDE that turns a prompt into a spec, then into a diff.

Free · Free Verified

50 credits a month on the perpetual free tier, with Claude Sonnet 4.5 and open-weight models, subject to rate limits. Add-on credits are not available. Kiro Web is not on this plan.

50 credits a month. Claude Sonnet 4.5 and open-weight models such as Qwen3 Coder Next, DeepSeek 3.2, and MiniMax M2.1, with rate limits. No add-on credits. Kiro Web is not included.

The 50-credit free allowance is on the pricing page and the billing docs. The free tier is not offered on Kiro Enterprise or in AWS GovCloud.

Enough for. A short look at a spec-driven IDE, if 50 published credits is enough to see whether you can follow the diff.

When free dies. When the 50 credits are gone. They return next billing cycle. Add-on credits are not sold on this plan, and Kiro Web is not included. Reset: Plan credits renew at the start of the next billing cycle. Paid subscriptions are processed on the 1st of each calendar month. Unused plan credits do not roll over. Add-on credits, which are paid plans only, last 12 months.

What to do when Kiro runs out

OpenAI Codex

Free-tier weight 2/5 · checked 2026-10-19

A coding agent included with ChatGPT, not a separate app subscription.

Free · Free Verified

Free includes a limited Codex trial. OpenAI does not publish a fixed prompt count. It is enough to see the product, not to build through the month.

Limited trial access to Codex and limited access to GPT-5.6. Codex pricing docs also list GPT-6 Luna at Standard speed in the desktop app, subject to rollout. No published prompt count.

Confirmed on the Codex pricing page as a limited trial. GPT-6 Luna on Free is listed as subject to rollout, so it is not treated as a guarantee.

Vendor does not publish a fixed count.

Enough for. Seeing what Codex is, if you already have a ChatGPT account. It is not a month of project work.

When free dies. When the limited trial ends. The vendor does not publish a fixed prompt count. Go sits above Free. A Codex pricing doc lists $8/mo, and chatgpt.com/pricing still does not print a dollar amount, so that row stays verify. Reset: Plan allowance, with weekly limits that may also apply. Local messages and cloud chats share it. OpenAI does not publish one universal reset clock.

What to do when OpenAI Codex runs out

Windsurf

Free-tier weight 2/5 · checked 2026-10-19

The editor formerly called Windsurf, now sold as Devin Desktop.

Free · Free Verified

A light agent quota, limited model access, unlimited inline edits, and unlimited Tab. Cognition does not publish the quota as a number.

Light agent quota, limited models, unlimited inline edits, and unlimited Tab. Quota size is not published.

Price and qualitative limits are from the Devin pricing page. The missing number is stated, not invented.

Vendor does not publish a fixed count.

Enough for. A light look at the editor now sold as Devin Desktop, with unlimited inline edits and Tab, if you can work in an IDE.

When free dies. When the light agent quota is gone. The vendor does not publish that quota as a number. Free cannot buy on-demand credits. Reset: Pro has daily and weekly quotas. Max has a weekly quota and no daily cap. Cognition does not publish the size. Past the quota, paid plans can buy extra usage at API pricing.

What to do when Windsurf runs out

The free plan does not include the tool

A $0 row can still appear because the vendor has a free chat plan. That is not a free coding tool.

Claude Code

Free-tier weight 0/5 · checked 2026-10-19

A terminal coding agent that shares your Claude plan.

Free · Free Verified

Free Claude does not include Claude Code. You need Pro or Max before this tool exists for you.

Claude chat only. Claude Code is not included.

Do not plan a coding workflow on the free plan.

Enough for. Nothing in this product. Free Claude is chat. Claude Code is not on that plan.

When free dies. Immediately, if you expected a coding agent. The tool starts on the paid floor on its page. Reset: A rolling five-hour window, plus weekly limits on paid plans. Chat and Claude Code share one pool.

What to do when Claude Code runs out

Where to go next

Questions people actually ask

What is the best free AI coding tool for a beginner?

For a web app you want to see running, Bolt.new is the free plan this catalog describes as one of the more usable trials: 1 million tokens per month and a 300,000 token daily cap. Public and private projects, Bolt branding, and a 10MB upload limit. Direct edits in Code view do not spend tokens. For an editor agent at $0, Google Antigravity is the individual plan with a weekly quota the vendor does not publish as a count. Claude Code is not a free coding tool.

Which free plans publish a real allowance, and which do not?

Lovable, Bolt, v0, Kiro, and Jules print a cap on the card. Kiro Free is 50 credits a month. Jules is 15 tasks per rolling 24 hours and 3 at once. Copilot prints two conflicting descriptions, so that row is marked verify. Cursor Free, Codex Free, Replit Starter, Windsurf, and Antigravity are labeled on the card when the vendor does not publish a fixed count. Amp Hobby publishes the price, which is $0, and does not include orb minutes or model credits. Claude Code's free row is chat only. Read the limit line on each card rather than a number repeated from memory.

What should I do when the free plan hits zero?

Check which meter ran out, wait if the reset is soon, drop to a cheaper mode if one exists, then upgrade or leave. That order is the limit playbook, with a section for every tool in this catalog.

Other guides

Take the quiz · When you hit a limit · How the weights work