Is Claude AI free?
Short answer: yes, four honest ways. Anthropic publishes three of them on claude.com. The fourth is the one most write-ups skip: Claude served through a hosted app where the host pays the per-token API fee on your behalf, so you use Claude with no Anthropic account at all and the price tag is $0. Below are the verified numbers for each path, including the line of code in mk0r that decides whose Anthropic bill an anonymous visitor's chat turn lands on.
Yes, Claude AI is free in four ways.
- claude.com free tier: Claude Sonnet 4.5 by default, roughly 15 to 40 messages per 5 hour rolling window. No credit card. No expiration.
- API trial credits: a small pool of credits on a new platform.claude.com account, enough for a few hundred Haiku calls.
- Mobile and desktop apps: Anthropic's iOS, Android, and desktop apps are free downloads with the same free-tier allowance.
- Hosted apps that pay the API fee: use Claude through a third-party host that routes inference through its own ANTHROPIC_API_KEY. No Anthropic account required. mk0r pins free sessions to Claude Haiku 4.5 and absorbs the cost.
Source: claude.com/pricing and platform.claude.com pricing docs.
The four free paths, ranked by friction
They are not the same thing. Each has a different trade between account setup, model access, and how long until you bump into a limit. Most pages on this question stop at the first three.
Where Claude AI is free in April 2026
- claude.com free tier. Sign in with email or Google. Sonnet 4.5 by default, roughly 15 to 40 messages per 5 hour rolling window. No credit card, no expiration.
- Free trial credits on the per-token API. New accounts at platform.claude.com get a small pool of credits, enough for a few hundred Haiku calls before you would need to load real money.
- Anthropic's free mobile apps (iOS + Android) and the free desktop app. Same usage allowance as the free tier on claude.com, just a different surface.
- Claude served through a hosted app that pays the per-token fee on your behalf. No Anthropic account at all. The host picks the model, the host eats the bill, you pay $0. mk0r is one example.
What the claude.com free tier gets you
Sign in at claude.com with an email or Google account. The default chat model is Claude Sonnet 4.5, the same intelligence powering most Pro interactions. You can upload documents, run web search, generate code, analyze images, and use memory across conversations.
The headline number that gets quoted is roughly 15 to 40 messages per 5 hour rolling window, but that range is fuzzy on purpose. The real allowance scales with how heavy your conversations are and how loaded the platform is at that moment. A short chat session of 10 prompts each averaging 200 tokens is much cheaper on the limit than a single 1500 token prompt with a large file attached.
What you do not get on the free tier: Claude Code (the terminal-based agent) which is gated to at least Pro, the longer 1M token context window on Sonnet and Opus, the projects feature, and priority routing during peak hours.
What the API trial credits actually buy you
Different surface, different bucket. Create an account at platform.claude.com and Anthropic deposits a small pool of credits in your balance. The dollar value is not published on the pricing page, so treat it as small. In practice, it is enough to try a few hundred Haiku calls or a few dozen Sonnet calls before you would need to put a real payment method on file.
Per-token rates as of April 29, 2026: Haiku 4.5 is $1 input and $5 output per million tokens, Sonnet 4.5 and 4.6 are $3 input and $15 output, Opus 4.5, 4.6, and 4.7 are $5 input and $25 output. Cache writes are 1.25x base for the 5 minute cache or 2x for the 1 hour cache. Cache reads are 0.1x base. The Batch API is half off both directions.
Trial credits do not auto-refill and they expire if the account stays inactive long enough that Anthropic closes the trial. Once they are gone, the API requires a top-up.
The fourth path: hosted apps that absorb the per-token cost
This is the one almost no other write-up walks through. A lot of the Claude on the web today is being served through hosted third-party apps that route every inference call through their own ANTHROPIC_API_KEY. From the visitor's side, there is no account, no key, no credit card. The host pays the per-token fee. The visitor pays $0.
mk0r is one example. The whole flow takes four steps when you land on the page.
What happens when you load mk0r.com for the first time
Open the page
No email, no Google sign-in, no API key. The auth provider runs on first paint.
Anonymous Firebase user
src/components/auth-provider.tsx line 68 calls signInAnonymously(auth). You now have a uid with no PII attached.
Free model pinned
src/app/api/chat/model/route.ts pins FREE_MODEL = "haiku" for any session that is not entitled to a paid model.
Host's API key, not yours
src/app/api/chat/route.ts line 187 selects the shared ANTHROPIC_API_KEY because the visitor has no OAuth token. Inference bills against mk0r's bucket.
The fork between “host pays” and “you pay” sits on a single line in src/app/api/chat/route.ts. The branch below is the actual code, side by side for the two cases. The apiKey variable is what gets passed to the inference call. When it resolves to the shared key, your tokens are on the host. When it resolves to the empty string and an OAuth token is set, your tokens are on your own Claude Pro or Max plan.
The line that decides whose Anthropic bill your tokens land on
// src/app/api/chat/route.ts (lines 172-189)
const sharedApiKey = process.env.ANTHROPIC_API_KEY;
let oauthToken: string | undefined;
if (!authedUser.isAnonymous) {
// anonymous visitor, branch skipped
}
// apiKey resolves to the shared ANTHROPIC_API_KEY.
// Inference for this turn is metered against
// the host's Anthropic account, not the visitor's.
// Cost to visitor: $0.
const apiKey = oauthToken ? "" : sharedApiKey ?? "";
if (!oauthToken && !sharedApiKey) {
return Response.json(
{ error: "ANTHROPIC_API_KEY not configured" },
{ status: 503 }
);
}“An anonymous visitor's chat turn on mk0r runs Claude Haiku 4.5 against the shared ANTHROPIC_API_KEY in chat/route.ts:187. The host absorbs the per-token cost (Haiku 4.5 is $1 input and $5 output per million tokens). FREE_MODEL = haiku is pinned at chat/model/route.ts:5 so the picker can not escalate the unsigned session to Sonnet or Opus.”
src/app/api/chat/route.ts:172-189, src/app/api/chat/model/route.ts:5, src/components/auth-provider.tsx:68
Where free stops being enough
Honest answer: free is enough longer than people expect, but not forever. The three usual triggers for upgrading off the free tier:
- You start bumping into the 5 hour rolling limit on a working day. Two long Sonnet conversations with file uploads is often enough.
- You want Claude Code, the terminal-based agent. It is not on the free tier and is not in any hosted-app fourth-path either. Pro at $20 per month is the lowest tier that includes it.
- You want the 1M token context window on Sonnet 4.6 or Opus 4.7 for a long document or whole codebase. Free pins you to the standard window.
For most chat-driven workflows, none of those three trigger in a given week and the free tier is the right place to live. For heavier loops, the next step up is Pro at $17 to $20 per month depending on annual or monthly billing.
How to spot the fourth path on a hosted app
If you are trying to figure out whether a third-party app is actually free or just running a trial that will start charging later, three quick checks:
- Is sign-up required? If the chat works without an account, the host is paying for your inference. If you have to verify an email before the first message, treat it as a trial.
- Does it ask for an API key? Apps that take your own ANTHROPIC_API_KEY are bring-your-own-key, you pay every token. Apps that have no key field are spending the host's key.
- Is the model pinned? Hosts that absorb cost almost always pin to the cheapest model in the catalog. If you are stuck on Haiku or a small variant on the free tier and Sonnet shows up only behind a paywall, that is the host doing the unit economics.
Want to talk through which Claude path fits what you're building?
15 minutes on Cal. We will look at your typical session shape, estimate where the free tier breaks, and tell you whether the fourth path is enough for your use case.
Frequently asked questions
Is Claude AI free?
Yes. Anthropic offers a free tier on claude.com (Sonnet 4.5 default, roughly 15 to 40 messages per 5 hour window, no credit card required). Anthropic also gives new per-token API accounts a small pool of free trial credits. The mobile apps for iOS and Android and the desktop app are free as well. Separately, you can use Claude through any hosted app that pays the per-token API fee on your behalf, with no Anthropic account at all.
What does the free tier on claude.com get you specifically?
Claude Sonnet 4.5 as the default chat model, web search, file uploads, document analysis, image analysis, code generation, memory across conversations, and the desktop and mobile apps. Roughly 15 to 40 messages per 5 hour rolling window. The number is not a hard quota, it scales with conversation length and current capacity. The free tier does not include Claude Code (the terminal agent), the projects feature on claude.com, the longer 1M token context window on Sonnet and Opus, or the agentic computer-use surface.
How do free API trial credits work?
When you create a new account at platform.claude.com, Anthropic deposits a small balance of credits you can spend at the published per-token rates. The balance does not appear on the public pricing page, so treat the dollar value as undocumented and small. In practice it is enough to try Haiku for a few hundred calls or Sonnet for a few dozen. Credits expire if your account stays inactive long enough that Anthropic closes the trial, and they do not auto-refill. Once they are gone, the API requires a real payment method on file.
Can I use Claude AI without any account at all?
Not on claude.com directly, the free tier requires you to sign in with an email or Google. But yes, on hosted apps that route inference through their own API key. mk0r is one example. When you load mk0r.com, src/components/auth-provider.tsx line 68 signs you in anonymously through Firebase (no email, no PII), and the chat API at src/app/api/chat/route.ts line 187 picks the host's ANTHROPIC_API_KEY for your session. The model pinned to free sessions is Claude Haiku 4.5, configured at src/app/api/chat/model/route.ts line 5 as FREE_MODEL = "haiku". Every input and output token bills against the host's Anthropic account at $1 input and $5 output per million tokens, not yours.
Why would a hosted app eat the per-token cost for an anonymous visitor?
Conversion economics. Claude Haiku 4.5 is the cheapest model in the catalog, $1 per million input tokens and $5 per million output. A typical mk0r generation run is on the order of 10,000 input and 5,000 output tokens, so a few cents per session. The host absorbs that on free sessions in exchange for the chance that a fraction of users upgrade to a paid plan or connect their own Claude.com Pro plan via OAuth. The shared key is not literally free for the host, it shows up on Anthropic's bill at the published rates. It is just absorbed at a unit-economics level.
Are there hidden trade-offs to using Claude through a hosted app?
Three real ones. First, you do not pick the model, the host does. mk0r pins free sessions to Haiku 4.5 because the cost works out at that price point. Sonnet and Opus are gated behind the host's paid tier or behind a connected user OAuth token. Second, your prompts and responses pass through the host's infrastructure, not Anthropic directly, so the host's privacy and retention policy applies on top of Anthropic's. Third, you do not control rate limits. The host sets them. If they are getting hammered by a viral post, your free session may queue or fail.
When does the free tier on claude.com stop being enough?
Three usual triggers. You start hitting the 5 hour rolling limit on long sessions and want to keep working. You need Claude Code (the terminal-based agent) which is gated to Pro and Max. You need to use Sonnet's or Opus's 1M token context window for a long document or codebase, which is also gated. At that point Pro at $20 per month (or $17 on the annual plan) is the next step up. Max at $100 or $200 per month is for power users who want 5x or 20x the Pro allowance.
Can I use the free tier for commercial work?
Yes, Anthropic's terms permit commercial use on the free tier. The constraints are operational, not legal. You hit the 5 hour rolling limit faster on a busy workday, you do not get the Claude Code terminal agent, and you do not get the priority routing or longer context windows. If commercial work is core to what you are doing, the Pro plan at $20 per month exists exactly for that.
Does the free tier on claude.com see your conversations for training?
Anthropic's data policy distinguishes between free, Pro, Max, and Team. As of April 2026, free tier conversations may be used to improve Claude under the privacy and acceptable use policies, with opt-outs available in settings. Pro and Max have stricter handling. The full current text lives at https://www.anthropic.com/legal/privacy. If your prompts contain proprietary or sensitive content, read it before deciding which tier you want to be on.
What about Claude Code? Is that free?
Claude Code, the terminal-based agent, is not on the free tier. It needs at least Pro at $20 per month, Max at $100 or $200 per month, or per-token API credits with a payment method on file. There is no free version of Claude Code on claude.com. There are open-source agents like the ACP protocol that mk0r uses internally to drive Claude inside an E2B sandbox, but those still need an inference path (the hosted Pro plan or a metered API key) to actually call the model.