Claude Code does not run on the free tier. Pro at $20/month is where it starts. This guide covers what you get, where the limits are, and the one upgrade decision you will face once you start using it daily.
What Claude Code Pro Includes
One subscription covers everything: Claude Code in the terminal, Claude on the web, desktop app, and mobile. All activity draws from the same usage pool. There is no separate Claude Code subscription.
Pro adds five things over the free tier: Claude Code access, Projects for persistent context across conversations, priority access during high-traffic periods, early access to new features, and a 5x higher usage allocation. The free tier gets none of the first two and a smaller allocation of everything else.
Claude Code Terminal Access
Run Claude Code in your terminal with full tool access: file reads and writes, Bash commands, web search, and MCP server integrations. Not available on the free plan at all.
Projects
Persistent context across conversations with uploaded documents and shared instructions. Store codebase context, style guides, and architecture docs that Claude reads at the start of every session.
Priority Access
Pro users get queue priority during peak hours when Anthropic's servers are under load. Free users see throttling or degraded responses during high-traffic periods.
API key override
If an ANTHROPIC_API_KEY environment variable exists on your machine, Claude Code uses that key instead of your Pro subscription. This means API charges apply at per-token rates, separate from your $20 subscription. Remove or unset the variable if you want Claude Code to use your Pro plan.
Usage Limits Explained
Two limits apply to Pro: a 5-hour rolling window and a weekly cap. They work independently and most users only encounter the first one.
The 5-Hour Rolling Window
Pro allows approximately 45 messages per 5-hour window. The window starts when you send your first message and resets when you send a message after the 5 hours have elapsed. It does not reset on a fixed clock schedule. If you send a message at 9am and your next message is at 2:30pm (after the window expired), a new window starts at 2:30pm.
The 45-message figure is a rough estimate. Actual capacity varies based on message length, file attachments, and conversation history. Short messages consume less than long ones that include full file contents.
The Weekly Cap
Pro plans have a weekly usage limit that resets 7 days after your session starts. Anthropic reports fewer than 2% of Pro users hit this with Sonnet 4.6, so it is not something most developers encounter. It exists to catch extreme usage patterns, not normal development work.
| Free | Pro ($20/mo) | Max 5x ($100/mo) | Max 20x ($200/mo) | |
|---|---|---|---|---|
| Messages per 5-hour window | ~15 (chat only) | ~45 | ~225 | ~900 |
| Sonnet hours per week | Very limited | 40-80 hours | 140-280 hours | 240-480 hours |
| Opus flat-rate access | No | No | 15-35 hrs/week | 24-40 hrs/week |
| Claude Code access | No | Yes | Yes | Yes |
| Projects | No | Yes | Yes | Yes |
| Weekly cap (% affected) | N/A | <2% | <2% | <1% |
What Happens When You Hit a Limit
Claude Code pauses and shows a message indicating you have reached your limit. You have three options: wait for the 5-hour window to reset, enable extra usage (pay-as-you-go at API rates with explicit opt-in), or upgrade to Max. Extra usage is never charged automatically.
Checking which authentication Claude Code is using
# If this is set, Claude Code uses API billing, not your subscription
echo $ANTHROPIC_API_KEY
# To use your Pro/Max subscription instead, unset it
unset ANTHROPIC_API_KEY
# Verify the active session
claude --versionModel Access: Sonnet vs Opus
Pro includes Sonnet 4.6 as the default model. Sonnet handles the vast majority of coding tasks: refactoring, debugging, writing tests, generating code from specs. For most Claude Code workflows, Sonnet is the right model.
Opus 4.6 is the more capable model, with better performance on complex architectural reasoning, ambiguous requirements, and long-horizon planning. It is also slower and more expensive on the API ($15/$75 per million input/output tokens vs Sonnet's $3/$15).
Pro does not include flat-rate Opus access. You can access Opus through extra usage at standard API rates, but this means pay-as-you-go billing on top of your $20 subscription. For developers who need Opus regularly, the Max 5x plan at $100/month is a more economical choice than paying API rates for Opus on top of Pro.
Sonnet 4.6 (Included in Pro)
$3 input / $15 output per million tokens via API. Default Claude Code model. Handles most coding tasks including debugging, refactoring, test generation, and multi-file changes. Faster than Opus with a lower cost per task.
Opus 4.6 (Max plans only, flat-rate)
$15 input / $75 output per million tokens via API. Better for complex architectural decisions, long-horizon planning, ambiguous requirements, and tasks that require extended reasoning. 5x more expensive than Sonnet per token.
When does Opus actually matter?
For most coding tasks, Sonnet and Opus produce comparable results. Opus consistently outperforms on: decomposing large ambiguous projects into concrete tasks, debugging subtle logic errors in complex systems, and generating architecture designs from high-level requirements. If you find yourself re-prompting Sonnet multiple times on the same task, that is a signal the task might benefit from Opus.
Free vs Pro vs Max: Full Comparison
The free tier is a trial. Pro is the working developer plan. Max is for developers who run Claude Code as their primary tool throughout the workday.
| Feature | Free | Pro ($20/mo) | Max 5x ($100/mo) | Max 20x ($200/mo) |
|---|---|---|---|---|
| Claude Code (terminal) | No | Yes | Yes | Yes |
| Claude app (web/mobile) | Yes (limited) | Yes | Yes | Yes |
| Projects | No | Yes | Yes | Yes |
| Sonnet 4.6 | Yes (limited) | Yes | Yes | Yes |
| Opus 4.6 (flat-rate) | No | No | Yes (15-35 hrs/wk) | Yes (24-40 hrs/wk) |
| Priority access | No | Yes | Yes (higher) | Yes (highest) |
| Extra usage available | No | Yes | Yes | Yes |
| API access included | No | No | No | No |
| Extended thinking | No | Yes | Yes | Yes |
API access is separate from all plans
None of the subscription plans (Free, Pro, Max) include API access for building applications. If you need to call Claude's API programmatically, you need a separate API key with per-token billing through the Anthropic Console. Subscriptions cover Claude Code and the Claude app only.
When to Upgrade to Max
The signals are clear if you pay attention. Most developers know within a week of using Pro whether they need Max.
Stay on Pro if...
You do 2-4 focused Claude Code sessions per week, each lasting 1-2 hours. You rarely hit the 5-hour window limit. You do not need Opus for your workflows. Your Claude Code use is supplemental to manual coding.
Upgrade to Max 5x ($100/mo) if...
You hit the 5-hour window limit on active coding days. You want Opus access for complex architectural work without per-token charges. You run parallel subagents. You use Claude Code for 4+ hours per day regularly.
Upgrade to Max 20x ($200/mo) if...
You run Claude Code continuously for a full workday. You use multi-agent orchestration workflows with 3+ parallel agents. Large codebases requiring extended context are common in your work. Claude Code is your primary development environment, not a supplement.
Consider API instead if...
Your usage is irregular (some weeks heavy, some weeks none). You are building applications that call Claude programmatically. Extra usage charges on Pro would be cheaper than Max given your actual monthly usage. You want per-task cost visibility.
The math on Max vs API: Sonnet API pricing is $3/$15 per million tokens. A typical Claude Code coding turn uses 20-30K input tokens and 3-5K output tokens. That is roughly $0.13-0.17 per turn. At 45 turns per 5-hour window, that is about $6-7.50 per session. If you do 3+ full sessions per week, Max 5x at $100/month is cheaper than API.
Cost Comparison vs Competitors
Three tools dominate the AI coding market at similar price points. They are fundamentally different products, not equivalent options at different prices.
| GitHub Copilot Pro ($10/mo) | Cursor Pro ($20/mo) | Claude Code Pro ($20/mo) | |
|---|---|---|---|
| Product type | IDE plugin | AI-native IDE | Terminal agent |
| Autonomy level | Suggestion-based | Agent mode available | Fully autonomous |
| Model access | GPT-5.4, o3-mini, Claude 3.5 | Multiple (Sonnet, GPT-5.4) | Sonnet 4.6 |
| Included requests/mo | 500 premium + unlimited slow | 500 fast requests | ~225 msgs/day (Pro limits) |
| Terminal/shell integration | No | Via composer | Native |
| Full codebase awareness | Limited | Yes | Yes |
| Autonomous commands | No | Limited | Yes |
GitHub Copilot at $10/month is the right choice for developers who want inline completions and occasional chat. Cursor at $20/month is the right choice for developers who want an AI-enhanced IDE. Claude Code Pro at $20/month is the right choice for developers who want an autonomous agent that operates in their terminal alongside their existing editor.
Many developers run Claude Code and Cursor simultaneously: Claude Code for autonomous multi-file tasks, Cursor for inline editing in the IDE. The two tools complement each other.
Tips to Get More from Pro
Pro's limits are per-turn, not per-token-of-useful-work. The biggest waste on Pro is turns that repeat work Claude already did. These patterns reduce unnecessary turns:
Use Projects for persistent context
Store your codebase architecture, coding conventions, and key file paths in a Project. Claude reads this context at session start. Eliminates the first 3-5 turns every session where you re-explain the codebase structure.
Batch related tasks in one session
Each session starts a 5-hour window. Running one complex task per session wastes window time. Group related tasks: refactor the auth module AND write tests AND update the docs in a single session.
Use CLAUDE.md for project context
Claude Code reads CLAUDE.md from your project root at every session. Put architecture diagrams, stack decisions, file ownership rules, and build commands there. Claude skips the exploration phase and starts working immediately.
Switch to Haiku for simple tasks
Use the /model command to switch to Haiku 4.5 for tasks like writing docstrings, formatting code, or generating boilerplate. Haiku runs faster, consumes fewer resources from your window, and handles simple tasks well.
CLAUDE.md that saves turns at session start
# Project: Auth Service
## Stack
- Node.js 22, TypeScript, Express 5
- PostgreSQL with Drizzle ORM (see src/db/schema.ts)
- Auth: JWT with jose library
## Key Files
- src/middleware/auth.ts — JWT validation middleware
- src/api/users.ts — User CRUD routes
- src/db/schema.ts — Full database schema
## Build Commands
- npm run dev — Start dev server on port 3000
- npm run test — Run all Jest tests
- npm run lint — ESLint check
## Conventions
- Errors use AppError class from src/lib/errors.ts
- All DB queries go through src/db/queries/ directory
- Tests mirror src/ structure in tests/A well-written CLAUDE.md eliminates the codebase exploration phase. Claude Code without CLAUDE.md spends 3-5 turns reading files to understand the project. With CLAUDE.md, it starts from context and uses those turns for actual work.
Frequently Asked Questions
What does Claude Code Pro include?
$20/month covers Claude Code in the terminal, Claude on web, desktop, and mobile, Projects for persistent context, priority access during high-traffic periods, and approximately 45 messages per 5-hour window on Sonnet 4.6. Flat-rate Opus access is not included. API access for building applications is also not included.
Does Claude Code work on the free plan?
No. Claude Code requires a Pro plan ($20/month) or a separate API key. The free tier only covers the Claude.ai web and app interface with approximately 40 short messages per day. There is no free trial of Claude Code.
What is the usage limit on Claude Code Pro?
Approximately 45 messages per 5-hour rolling window. The window starts with your first message and resets when you send a message after 5 hours expire. There is also a weekly cap, but Anthropic reports fewer than 2% of Pro users hit it with Sonnet 4.6. Extra usage is available at API rates ($3/$15 per million Sonnet tokens) with explicit opt-in.
Can I use Opus with Claude Code Pro?
Not in the flat-rate. Opus is available through the extra usage system at standard API rates ($15 input / $75 output per million tokens), but this is pay-as-you-go on top of your $20 subscription. For regular Opus access without per-token charges, upgrade to Max 5x ($100/month, includes 15-35 Opus hours per week) or Max 20x ($200/month, includes 24-40 Opus hours per week).
Is usage shared between the Claude app and Claude Code?
Yes. All activity in the Claude web app, desktop app, mobile app, and Claude Code terminal draws from the same 5-hour window allocation. Heavy app use reduces how many Claude Code turns you get in that window, and vice versa.
When should I upgrade from Pro to Max?
Upgrade to Max 5x when you regularly hit the 5-hour window limit on active coding days, want flat-rate Opus access, or run parallel subagents. Upgrade to Max 20x when Claude Code runs for 6+ hours daily, you use multi-agent orchestration workflows, or you work on large codebases continuously. Stay on Pro if you do 2-4 focused sessions per week and rarely hit limits.
How does Claude Code Pro compare to Cursor?
Both cost $20/month. Cursor is an AI-native IDE with inline completions and an agent mode. Claude Code is a terminal agent that operates autonomously across your full codebase without requiring an IDE. Many developers run both: Claude Code for autonomous multi-file tasks, Cursor for inline editing. They are complementary, not competing.
Does Claude Code Pro include API access?
No. None of the Claude subscription plans (Free, Pro, Max) include API access for building applications. API access requires a separate setup through the Anthropic Console with per-token billing. If you have an ANTHROPIC_API_KEY environment variable set, Claude Code uses that key (and API billing) instead of your subscription.
Claude Code Generates More Code. Morph Applies It Faster.
Claude Code Pro sessions produce large diffs across multiple files. Morph's Fast Apply model merges LLM-generated edits deterministically at 10,500+ tokens per second. The reliability layer between your Claude Code sessions and your codebase.