Claude Code Pricing At A Glance
Claude Code is bundled with three consumer subscription tiers and one team tier. You can also bypass subscriptions entirely and pay per token via the Anthropic API. Pick the row that matches your usage pattern.
| Plan | Monthly Price | Claude Code | Best For |
|---|---|---|---|
| Free | $0 | No | Web chat only |
| Pro | $20 (or $17 annual) | Yes | Hobbyist, under 2 hrs/day |
| Max 5x | $100 | Yes | Daily developer, 4-6 hrs/day |
| Max 20x | $200 | Yes | Heavy use, parallel agents |
| Team Standard | $20/seat | No | Chat-only teams |
| Team Premium | $100/seat | Yes | Dev teams |
| API pay-per-token | $0 base + usage | Yes (via OAuth) | Automation, CI, cron jobs |
According to Anthropic's pricing page, Pro drops to $17/month with annual billing ($204/year). Max tiers have no annual discount.
Claude Pro ($20): The Entry Point
Pro gives you roughly 45 messages every 5 hours across chat and Claude Code combined. This is fine for a few hours of coding per day, but long tool-heavy sessions can hit the rate limit in 30-40 minutes.
What you get
- Claude Code CLI access
- Sonnet 4.6 default model
- Limited Opus 4.7 (fair-use)
- Projects, artifacts, files
What bites
- ~45 messages / 5 hours
- Weekly Opus cap (hours)
- Rate-limited on long sessions
- Single concurrent session
In April 2026, Anthropic briefly removed Claude Code from Pro for new signups. After widespread backlash on X and Reddit, the change was reversed within 48 hours. Claude Code remains in the Pro tier as of this writing.
Max 5x ($100) vs Max 20x ($200): The Real Difference
Max tiers are sized by multiples of Pro. Max 5x gives 5x the Pro usage envelope. Max 20x gives 20x. The gap between them matters most if you leave Claude Code running in background agents or run parallel worktrees.
Max 5x
~225 messages per 5-hour window. Enough for 4-6 hours of sustained coding with Sonnet. Opus gets a weekly hour cap around 35-40 hours. Solo developers hit this sweet spot.
Max 20x
~900 messages per 5-hour window. Supports 2-4 concurrent Claude Code sessions, parallel worktree agents, or leaving a long-running agent loop running 24/7. Opus hours roughly double vs Max 5x.
The jump from Max 5x to Max 20x is 2x the price for 4x the usage. If you routinely hit the 5x limit before 2pm, Max 20x is the better value. If you only hit it occasionally, stay on 5x and grind through the reset.
The complete playbook
Stop guessing at Claude Code costs. Get the full guide with model routing configs, cache strategies, and the exact setups we use to run under $30/month.
This post covers what each plan costs. The guide covers how to pick the right plan for your workload, set up model routing to cut 60% of API bills, and the hidden gotchas that silently burn tokens.
Get the KaiShips Guide to Claude Code -- $29API Pay-Per-Token Pricing (April 2026)
If you run Claude Code in CI pipelines, cron jobs, or headless automation, pay-as-you-go API access is usually cheaper than a Max plan. Opus 4.7, released April 16, 2026, dropped Opus pricing by 3x from previous 4.x versions.
| Model | Input $/MTok | Output $/MTok | Context |
|---|---|---|---|
| Claude Opus 4.7 | $5.00 | $25.00 | 1M tokens |
| Claude Sonnet 4.6 | $3.00 | $15.00 | 1M (beta) |
| Claude Haiku 4.5 | $1.00 | $5.00 | 200K tokens |
All three tiers share a 5x output-to-input ratio. Output tokens are always 5x more expensive than input tokens, which is why verbose responses balloon bills faster than large contexts do.
Discount Mechanisms
Prompt caching: up to 90% off input tokens
Cache reads cost ~10% of standard input rate. Cache writes cost 25% more than input. Break-even after 2 reuses. Ideal for CLAUDE.md, tool definitions, and stable system prompts.
Batch API: 50% flat discount
Submit async jobs that return within 24 hours. Perfect for overnight data processing, bulk code review, or documentation generation. Not usable for interactive Claude Code sessions.
Amazon Bedrock / Google Vertex: competitive rates
Claude models on Bedrock and Vertex AI match Anthropic API list prices. The value is enterprise billing consolidation, VPC isolation, and regional data residency rather than outright discount.
Hidden Costs: Why Claude Code Bills Explode
Claude Code burns tokens differently from web chat. A 200-turn coding session can cost 10x per turn compared to turn 1, purely because every turn carries more accumulated context. Here are the four silent budget killers.
Context resubmission loops
Every tool call, file read, and sub-task replays the full conversation context. A single complex event can consume 50,000 to 300,000 tokens.
Auto-compact events
When a session nears the context limit, Claude Code triggers auto-compact, which summarizes prior context. Each compaction costs 100-200K tokens in a single call, usually at the Sonnet or Opus rate.
Per-turn cost drift
Longer sessions grow per-turn cost roughly linearly. At turn 200, each prompt costs around 10x what the same prompt cost at turn 1. /clear between tasks avoids this.
Subagent and tool overhead
Each subagent spawn pays its own bootstrap cost. Tool results get piped back into the parent session context. Heavy MCP server use adds 5-15% overhead.
Real Monthly Cost Scenarios
Theoretical rates only matter if you can map them to real workloads. Here are four profiles with observed monthly spend, based on user reports and our own KaiShips agent running 24/7 on Claude Code.
$20/mo
2 hrs/day, side project, mostly Sonnet, rarely hits limits. Pro plan covers it comfortably.
$100/mo
5 hrs/day, mix of Sonnet and Opus, occasional parallel worktrees. Max 5x fits; Pro would hit limits by noon.
$200/mo
8+ hrs/day, Opus-heavy, 2-4 concurrent sessions, leaves agents running overnight. Max 20x or API with cache.
$30-600/mo
Cron jobs, CI, headless. API pay-per-token with caching and model routing. KaiShips agent runs ~$30/mo tuned.
Which Plan Should You Pick?
Use this decision flow. Start at the top. Stop at the first yes.
- ●Running headless / cron / CI? Skip subscriptions. Use API pay-per-token with prompt caching enabled.
- ●Team of 5+ devs? Team Premium at $100/seat. Avoid Team Standard ($20/seat) -- no Claude Code access.
- ●Coding 8+ hrs/day or running parallel agents? Max 20x at $200/month.
- ●Coding 4-6 hrs/day with occasional Opus use? Max 5x at $100/month. Sweet spot for solo developers.
- ●Side project, under 2 hrs/day? Pro at $20/month. Annual billing drops to $17/month.
- ●Not sure yet? Start with Pro. Upgrade to Max 5x when you start hitting rate limits weekly. Prorated billing makes switching cheap.
Five Ways to Cut Your Claude Code Bill
These apply whether you pay via subscription (to stop hitting rate limits) or API (to lower invoices). Every strategy is stackable.
Enable prompt caching
Add cache_control markers around your system prompt and CLAUDE.md. Cache reads cost 10% of normal input. Biggest single lever for heavy users.
Route models by task
Default to Sonnet. Use Haiku for status checks, file lookups, and simple edits. Reserve Opus for complex refactors or multi-file reasoning.
Use Batch API for non-urgent work
Overnight code reviews, documentation generation, and bulk migrations. Submit a batch, walk away, come back at half price.
Trim CLAUDE.md under 4KB
Every API call loads your CLAUDE.md. A 20KB file adds 5,000 tokens to every request. Keep it lean and push details into skill files loaded on demand.
Run /clear between tasks
Every unrelated turn you keep in the session pays again. /clear resets context when switching focus. Prevents the turn-200 cost drift.
Related Reading
Deeper walkthroughs on the Claude Code features that actually shape your monthly bill.
Best Claude Code Configuration
Model routing, hooks, MCP servers, and the settings that keep costs predictable.
Claude Code Memory System
The 3-layer memory structure that keeps context small and API calls cheap.
CLAUDE.md Setup Guide
Keep your CLAUDE.md lean to avoid loading 5,000+ tokens on every single API call.
Claude Code vs Cursor
How pricing compares between Claude Code and Cursor for the same daily workload.
Claude Code Pricing FAQ
The most common questions we get about Claude Code billing, plan selection, and cost control.
How much does Claude Code cost per month?
Claude Code ranges from $0 to $200+/month. It is included in Claude Pro ($20/month), Max 5x ($100/month), Max 20x ($200/month), and all Team Premium and Enterprise plans. Pay-as-you-go API access charges per token: roughly $3/MTok input for Sonnet 4.6. Heavy solo developers typically pick Max 5x ($100) for predictable usage.
Is Claude Code included in the Claude Pro $20 plan?
Yes, Claude Code is included in the Claude Pro $20/month plan as of April 2026. Anthropic briefly removed Claude Code from Pro in early 2026, then reversed the change after user backlash. Pro gives roughly 45 messages every 5 hours across chat and Claude Code combined, which is enough for light hobbyist use.
What does the Claude API cost per million tokens?
As of April 2026, Claude Haiku 4.5 costs $1 input and $5 output per million tokens. Sonnet 4.6 costs $3 input and $15 output. Opus 4.7 costs $5 input and $25 output. Opus 4.7 pricing dropped from the previous $15/$75 rate, making premium reasoning 3x cheaper than Opus 4.5 was.
What is the difference between Max 5x and Max 20x?
Max 5x at $100/month gives 5x the usage of Pro -- roughly 225 messages per 5-hour window. Max 20x at $200/month gives 20x Pro usage, around 900 messages per 5-hour window. Max 20x also gets higher weekly Opus hours. Pick 5x if you code 4-6 hours/day, pick 20x if you run multiple parallel agents or leave Claude Code running all day.
Does Team Standard include Claude Code?
No. Team Standard at $20/seat/month does NOT include Claude Code access. Only Team Premium at $100/seat/month includes Claude Code. This is the single most common billing mistake -- organizations upgrade from Pro to Team Standard expecting feature parity and lose Claude Code entirely. Pick Team Premium or keep individual Pro/Max seats.
How can I reduce Claude Code token costs?
Five high-impact moves: enable prompt caching (saves up to 90% on repeated context), use Sonnet as default instead of Opus (5x cheaper), route simple tasks to Haiku (5x cheaper than Sonnet), trim CLAUDE.md under 4KB (cuts system prompt tokens), and use Batch API for non-urgent jobs (flat 50% discount).
Why does Claude Code burn more tokens than the chat app?
Claude Code resubmits full file contexts on every turn, loads system prompts and CLAUDE.md files on each request, and auto-compacts long sessions at 100-200K tokens per compaction. A 200-turn session can cost 10x per turn compared to turn 1. Chat workloads stay in a simpler conversation loop without file reads or tool use.
The complete playbook
Stop guessing at Claude Code costs. Get the full guide with model routing configs, cache strategies, and the exact setups we use to run under $30/month.
This post covers what each plan costs. The guide covers how to pick the right plan for your workload, set up model routing to cut 60% of API bills, and the hidden gotchas that silently burn tokens.
Get the KaiShips Guide to Claude Code -- $29