The Claude free plan gives you chat with a session-based usage limit that resets every five hours, plus basic code generation, file creation, and web search. It does not include Claude Code, and it cannot buy extra usage, because usage credits start at Pro. Anthropic describes Pro as at least five times the free allowance per session, which is the only published anchor for how large the free limit is. Any agent coding work needs Pro at $20 a month or an API key.
- Free gives you Claude chat with a session limit that resets every five hours.
- Claude Code is not included, at any usage level. This is a hard product boundary, not a throttle.
- The limit is not a message count. It tracks tokens, so long attachments cost far more than short questions.
- Pro is described as at least 5x free usage per session, the only published sizing Anthropic gives.
- Usage credits do not apply to Free. They start at Pro, so there is nothing to top up.
- The free route to real agent coding is an API key with a budget limit, or Gemini CLI.
What the free plan includes
Free against Pro, checked 7 August 2026.
| Free | Pro ($20/mo, $17/mo annually) | |
|---|---|---|
| Claude chat | Yes, session-capped | Yes, at least 5x free per session |
| Claude Code | No | Yes |
| Cowork | No | Yes |
| Basic code generation and file creation | Yes | Yes |
| Web search | Yes | Yes |
| Model selection | Limited | All models |
| Usage window | Session limit, resets every 5 hours | Session limit plus a weekly limit |
| Buy extra usage | No | Yes, usage credits at API rates |
| Priority access at peak times | No | Yes |
How the free limit actually behaves
Free uses the same shape as every paid plan: a session-based usage limit that resets every five hours. There is no separate weekly cap because there is no weekly allowance to cap.
Anthropic publishes no message number for it, and the number genuinely moves. What consumes it, in the order that matters:
- Message length. Consumption tracks tokens, so a long pasted document counts for far more than a short question.
- File attachment size. Attachments are part of the message and are re-sent as the conversation continues.
- Conversation length. Every turn carries the whole thread. The tenth message in a thread costs more than the first.
- Tool usage. Research and web search pull results into context, and those results are input tokens.
- Model and effort level. A bigger model, or more reasoning, consumes the allowance faster for identical work.
How far it goes in practice: short questions get you a comfortable number in a five-hour window. Pasting a 20-page document and iterating on it gets you noticeably fewer. If you are hitting the cap constantly on short prompts, you are past what the free tier is designed for.
Three genuinely free ways to code with an agent
If the goal is to write code with an AI agent and pay nothing, the Claude free plan is the wrong tool. These three are the real options as of August 2026.
Free tiers that cover agent coding, checked 7 August 2026. Terms move; verify before relying on any of them.
| Tool | Free allowance | What it actually is |
|---|---|---|
| Gemini CLI | 1,000 model requests per user per day with a personal Google account | A real terminal agent. The closest free equivalent to Claude Code |
| GitHub Copilot Free | 2,000 completions per month | Editor autocomplete and inline edits, not a terminal agent |
| Cursor Hobby | Free tier, no card required | An AI editor with limited agent usage |
| Claude free plan | Chat only, 5-hour session limit | Not an agent. Paste code in, copy code out |
The cheapest paid way to use Claude Code
An API key is the right tool for trying Claude Code without a monthly commitment. It gives you the real product with no usage windows, and you stop paying the moment you stop using it.
What a small API balance buys on Sonnet 5 at standard rates of $3 in and $15 out per million tokens.
| Spend | Roughly |
|---|---|
| $1 | 6 to 10 small focused fixes on code you can point at |
| $5 | An evening of real work, or about five twenty-turn agent sessions |
| $20 | Comparable to a light month, with no windows and no weekly cap |
Create the key in a workspace you can cap
Do not use a default personal workspace for agent work. You want the spend isolated so a budget limit applies to it specifically.
Set the budget limit before you use it
Console budget limits and alerts are the only thing standing between a runaway loop and a large invoice. Set both a warning threshold and a hard limit.
Point Claude Code at it
export ANTHROPIC_API_KEY="sk-ant-..."
claudeSwitch to a subscription if you keep using it
Interactive daily use on a key costs several times what Pro or Max does. The key is the right way to try Claude Code and the wrong way to live in it.
When free is genuinely enough
- Occasional questions, explanations, and short reviews of code you paste in yourself.
- Learning and study, where the volume is naturally low and the threads are short.
- Deciding whether you like the model before paying for anything.
- One-off writing and analysis that fits inside a five-hour window.
It stops being enough the moment you want the model to read your repo, run your tests, or edit files, because those are Claude Code, and Claude Code is not on the free plan.
Questions people ask
Yes, there is a real free tier for Claude chat with a session-based usage limit that resets every five hours, plus basic code generation, file creation, and web search. It does not include Claude Code or Cowork.
There is no fixed number. The limit tracks tokens rather than messages, so message length, attachment size, conversation length, tool use, and model choice all change how many you get. Anthropic describes Pro as at least five times the free allowance per session, which is the only published sizing.
No. Claude Code requires a paid plan from Pro upward, or an API key that bills per token. The API route lets you try it for a few dollars rather than a monthly fee, and a console budget limit keeps it bounded.
The free plan uses a session-based usage limit that resets every five hours. There is no weekly cap on Free, because the weekly limits apply to paid plans.
No. Usage credits, which let you continue past your included allowance at standard API rates, are available on Pro, Max 5x, and Max 20x, and on Team and Enterprise when an admin turns them on. Free has nothing to top up.
An API key with a small balance and a console budget limit. It gives you real Claude Code with no windows, and you pay only for what you use. If you use it most days, switch to Pro at $20 a month, because interactive daily use on a key costs several times more.
Gemini CLI. It is a genuine terminal agent that edits files and runs commands, with 1,000 model requests per user per day free on a personal Google account. GitHub Copilot Free is an editor tool with 2,000 completions a month, which is a different category.
If you write code, yes, because it is the only subscription route to Claude Code, and it drops to about $17 a month billed annually. For chat alone the free tier is genuinely usable and Pro mainly buys you headroom and priority at peak times.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.