Claude free plan: what you actually get
Exactly what the free Claude plan includes in August 2026 - web search, memory, file creation, connectors and Projects - and how its usage limits work.
The Claude free plan gives you the full chat app on web, iOS, Android and desktop - plus web search, memory, file creation, code execution, Skills, extended thinking, Google Workspace connectors, one custom MCP connector and up to five Projects. It is metered, not crippled. What it does not include is Claude Code, Cowork, Research, chat search, model selection and the 1M-token context window.
Most of what is written about Claude Free online is out of date or simply invented. This page separates what Anthropic actually documents from what people repeat, as verified on 21 August 2026.
#Claude Free at a glance
- Price
- $0, no card required
- Where it runs
- Web, iOS, Android, ChromeOS, macOS, Windows, Linux beta
- Projects
- Up to 5, without RAG project knowledge
- Custom connectors
- 1
- Published message limit
- None - Anthropic publishes no absolute count
- Commercial use of outputs
- Permitted, same as paid plans
#What does the Claude free plan actually include?
More than most comparison articles credit it with. Anthropic's pricing page lists chat on web, iOS, Android and desktop, code generation, data visualisation, web search, memory, file creation, code execution, desktop extensions, connectors with remote MCP, and extended thinking. Help-centre articles add several more.
| Capability | On Free? | The detail that matters |
|---|---|---|
| Web search | Yes | Cited answers, same as paid. It is the prerequisite for Research, which Free lacks. |
| Memory | Yes | Documented as available on Free, Pro and Max. Builds standing notes across chats. |
| File creation | Yes | Real .xlsx, .docx, .pptx and .pdf files, built by Anthropic's document Skills. |
| Code execution | Yes | The same toggle that gates Artifacts and Skills. On by default. |
| Artifacts | Yes | Side-panel documents, code, SVGs, React components. Persistent storage needs Pro. |
| Skills | Yes | Including custom Skills you upload yourself under Customize → Skills. |
| Projects | Capped at 5 | No RAG-enhanced project knowledge, so a huge knowledge base is not retrieved intelligently. |
| Custom connectors (remote MCP) | 1 | Free accounts are limited to a single custom connector. |
| Google Workspace connectors | Yes | Drive, Gmail and Calendar, with direct user authentication. |
| Extended thinking | Yes | Toggled from the tools menu on supported models. |
| Voice mode | Yes (beta) | Two-way spoken conversation on mobile, desktop and web. All plans. |
| Incognito chats | Yes | Excluded from memory, chat search and any training use. |
| Commercial use of outputs | Yes | Anthropic's Consumer Terms assign you its rights in outputs. See commercial use rights. |
#Myth one: "the free plan is browser-only"
It is not. A free account signs in to the official iOS app, the official Android and ChromeOS app, and the desktop apps for macOS 11+, Windows 10+ and the Linux beta. All download routes are on the download page.
What is plan-gated inside those apps are the extra tabs. The desktop app's Claude Code and Cowork panels require a paid plan; chat itself does not. On mobile, the iOS “Ask Claude” App Intent, Siri integration, widgets and the Analyze Photo control all work on Free - and all count against the same usage allocation.
#Myth two: the free plan has a fixed message allowance
Anthropic publishes no absolute message count for any plan - not for Free, not for Pro, not for Max. The frequently repeated “about 45 messages per five hours” figure appears nowhere in current documentation. What Anthropic publishes are relative statements: Pro offers “at least five times the usage per session” compared with the free service, Max is 5× or 20× Pro, Team standard seats 1.25× and premium seats 6.25×. Anything more precise than that is somebody's estimate presented as a fact.
#So how do the free limits actually work?
Three mechanics, all documented.
#1. A rolling five-hour session window
Your session allowance resets every five hours on a rolling basis. Hit it at 2pm and you are back in business around 7pm - you are not locked out for the day.
#2. A weekly cap sitting on top
Separately, there is a weekly limit across all models that resets on a fixed day and time assigned to your account, not on a calendar week that is the same for everyone. Anthropic also reserves the right to apply monthly caps or other restrictions at its discretion. Opus models are tracked on their own separate weekly counter alongside the all-models counter.
#3. One shared pool across every surface
Chat on web, chat on mobile, chat on desktop - all one allocation. On paid plans the same pool also covers Claude Code, Cowork, Claude Design and Claude for Excel, which is why heavy agentic work eats a subscription so quickly.
What consumes it fastest is predictable: long conversations (the whole thread is re-read every turn), large attachments, high effort levels, and tool-heavy turns. What is free of charge against your limit: automatic context compaction on paid plans, and the cached portion of project knowledge you have already referenced. Watch the meter under Settings → Usage. The full mechanics, including what happens at the wall, are on the pricing page.
#What the free plan excludes
The exclusion list is where the real upgrade pressure lives.
- Claude Code - agentic coding in the terminal. Authentication requires Pro, Max, Team, Enterprise or a Console API key.
- Cowork - the terminal-free agentic surface for document and research work. All paid plans since April 2026.
- Research - multi-step agentic search with citations. Paid only, and it requires web search.
- Chat search - retrieving from your own past conversations. Paid only, and arguably the most-missed gap.
- Claude Design and Claude Science - the Labs preview and the research workbench.
- Claude in Chrome and Claude for Excel / Microsoft 365.
- Model selection - no picker. You get what Anthropic assigns, and it does not publish which model that is.
- Fable 5 - explicitly unavailable on Free.
- Usage credits - Free accounts cannot buy their way past a limit; only Pro and Max can.
- The 1M-token context window - cited as a paid-plan capability.
The complete row-by-row comparison across all five plans is in the feature and plan matrix.
#Free or Pro: which should you actually pay for?
Pro is $20 a month, or $200 a year billed up front. Use this to decide honestly.
| If this describes you | Free | Pro ($20/mo) |
|---|---|---|
| A few questions a day, mostly writing and explaining | Enough | Unnecessary |
| You keep hitting the five-hour wall mid-task | Frustrating | Worth it - at least 5× the session usage |
| You want to search your own conversation history | Not possible | Chat search |
| You want cited multi-step research reports | Web search only | Research |
| You write code and want an agent in your terminal | Not available | Claude Code included |
| You want to delegate long document and file tasks | Not available | Cowork included |
| You need more than five Projects, or real retrieval over a big knowledge base | Capped at 5, no RAG | Unlimited + RAG |
| You want to pick Opus 5 over Sonnet 5 for a hard problem | No picker | Model selection |
| You need to paste a 400-page document into one chat | Standard context | Up to 1M tokens |
| You need image or video generation | Neither. Claude generates text and code only, on every plan. See Claude vs ChatGPT. | |
The short version: pay for Pro when you want agents or volume. If you want a very good writing and reasoning assistant a few times a day, Free genuinely does that job, and the prompting techniques that improve output cost nothing.
#Cheaper than Pro: the pay-as-you-go route
There is no plan between $0 and $20 - a real gap in Anthropic's line-up, and one competitors have filled. If Free is not quite enough but $20 a month is more than your usage justifies, the Claude API bills per token instead of per month.
| Model | Input per million tokens | Output per million tokens |
|---|---|---|
| Haiku 4.5 | $1 | $5 |
| Sonnet 5 | $2 | $10 |
| Opus 5 | $5 | $25 |
For genuinely light use the arithmetic is favourable: a few hundred short exchanges a month on Sonnet 5 or Haiku 4.5 lands well under $20, and prompt caching (cache reads at 0.1× base input) plus the Batch API's flat 50% discount cut it further. Getting started takes one API key from platform.claude.com and about six lines of Python - the Sonnet API page has the request shape.
An API key is not a subscription. You get no claude.ai chat interface, no Projects, no Memory, no Artifacts, no connectors directory and no mobile app - you supply your own client, or use a third-party one. Costs are uncapped by default rather than metered, so set a spend limit in the Console. And a Pro subscription does not include API credit, nor the reverse.
#Where Claude Free loses
Stated plainly, because it affects the choice. ChatGPT Free and ChatGPT Go now offer unlimited text, ad-funded in the US, while Claude Free is metered. Anthropic has nothing between $0 and $20, whereas Google AI Plus and ChatGPT Go both sit in the $5–$8 band. And on every Claude plan there is no image, video, audio or music generation at all, and no video or audio input.
What Free still wins on: quality of long-form reasoning and writing, a genuinely useful set of tool features at $0 that most free tiers withhold, and unrestricted commercial use of what it produces. Head-to-head detail is on Claude vs Gemini.
#Frequently asked questions
How many messages do you get on the Claude free plan?
Anthropic does not publish a number. Free usage is governed by a rolling five-hour session window plus a weekly cap that resets on a fixed day assigned to your account. Only relative figures are published - Pro is described as at least five times the free session usage.
Can I use the Claude apps on the free plan?
Yes. The official iOS, Android, ChromeOS, macOS, Windows and Linux beta apps all sign in with a free account. Only the Claude Code and Cowork tabs inside the desktop app require a paid plan. The free tier is not browser-only, despite what many articles claim.
Does the free plan include web search and file uploads?
Yes to both, plus memory, code execution, file creation, Artifacts, Skills, extended thinking, voice mode, Google Workspace connectors and one custom MCP connector. Research - the multi-step agentic search product - is separate and requires a paid plan.
Can I use free Claude output for commercial work?
Yes. Anthropic's Consumer Terms assign users all of its right, title and interest, if any, in outputs, and the help centre confirms outputs can be used in products and for customers. The one universal prohibition is using outputs to train a competing AI model.
Is there a plan between Free and Pro?
No. Anthropic offers nothing between $0 and $20 per month as of August 2026. For light users the practical alternative is the pay-as-you-go API, where Haiku 4.5 costs $1 per million input tokens and Sonnet 5 costs $2 - but it comes with no chat app, Projects or memory.
Does Claude train on free-plan conversations?
Only if you opt in. Since the policy change of 28 August 2025, training on consumer chats is opt-in across Free, Pro and Max. Allowing it means five-year retention; declining means 30 days. Incognito chats are never used for training regardless of the setting.
Free-plan inclusions checked on 21 August 2026 against claude.com/pricing and the individual feature articles at support.claude.com. Sign up or upgrade at claude.ai - this site cannot create accounts or take payment, and Anthropic changes free-tier limits without notice.