Claude Code Pricing Explained: Every Plan and Model (2026)

Claude Code pricing has four distinct tiers plus direct API access, each with different token allowances, model access, and rate limits. Choosing the wrong plan wastes money – a heavy API user on the Max plan overpays, while a light user on API Direct pays per-token for capacity they could get flat-rate on Pro. This guide breaks down every option with real numbers so you can pick the right one. Use the Cost Calculator to estimate your specific monthly spend based on your actual usage patterns.

Plan Comparison Table

Feature Free Pro ($20/mo) Max Individual ($100/mo) Max Team ($200/mo/seat) API Direct
Claude Sonnet Yes Yes Yes Yes $3/$15 per MTok
Claude Opus No Limited Yes Yes $15/$75 per MTok
Claude Haiku Yes Yes Yes Yes $0.25/$1.25 per MTok
Rate limit Low Moderate High Highest Pay-per-use
Context window 200K 200K 200K 200K 200K
Extended thinking No Yes Yes Yes Yes
Max output tokens 4K 8K 16K 16K 16K
Headless/CI mode No No Yes Yes Yes

Token prices shown as input/output per million tokens

Free Tier

The free tier gives you access to Claude Sonnet and Haiku with restrictive rate limits. You get roughly 20-30 messages per day depending on complexity. No Opus access, no extended thinking, no headless mode.

Best for: Evaluating Claude Code before committing to a paid plan. Not viable for daily development work.

Pro Plan ($20/month)

Pro unlocks higher rate limits and limited Opus access. You can run approximately 200-300 Sonnet messages per day, or about 50 Opus messages. Extended thinking is available, which significantly improves performance on complex reasoning tasks.

Best for: Individual developers using Claude Code 1-2 hours per day for code review, small refactors, and question answering. If you never hit rate limits on Pro, you do not need Max.

# Check your current plan
claude config get plan

Max Individual ($100/month)

Max removes most rate limits for individual use. You get substantially higher Opus quotas, headless mode for CI/CD pipelines, and priority API access during peak times. The key unlock is headless mode – if you want Claude Code running in GitHub Actions or automated workflows, this is the minimum tier.

Best for: Professional developers using Claude Code 3+ hours per day, anyone running headless automation, and developers who frequently use Opus for complex tasks.

Max Team ($200/month per seat)

Team adds admin controls, usage dashboards, and SSO. The per-seat cost is higher but includes shared billing, usage attribution per team member, and organization-wide CLAUDE.md management.

Best for: Teams of 3+ developers where centralized billing and usage tracking matter. The breakeven versus individual Max is at 1 seat – the extra $100/mo buys admin features, not more tokens.

API Direct (Pay Per Token)

Skip the subscription entirely and pay only for tokens consumed. This requires an Anthropic API key and manual cost management. There is no monthly minimum.

Per-Model Token Pricing

Model Input (per 1M tokens) Output (per 1M tokens) Effective cost per message*
Claude Haiku $0.25 $1.25 $0.002 - $0.01
Claude Sonnet $3.00 $15.00 $0.02 - $0.15
Claude Opus $15.00 $75.00 $0.10 - $0.80

Effective cost per message assumes 500-2000 input tokens and 200-1000 output tokens per typical coding interaction.

When API Direct Beats Subscriptions

The breakeven math is straightforward:

  • Pro breakeven: ~$20/mo in API costs = ~1,300 Sonnet messages or ~130 Opus messages
  • Max breakeven: ~$100/mo in API costs = ~6,500 Sonnet messages or ~650 Opus messages

If you consistently use fewer messages than these thresholds, API Direct saves money. If you exceed them, the subscription is cheaper.

# Set up API Direct access
export ANTHROPIC_API_KEY="sk-ant-your-key"
claude --model claude-sonnet-4-20250514

Hidden Cost Factors

Several factors inflate your actual costs beyond the base token prices:

System prompt tokens: Your CLAUDE.md file is sent with every message. A 3,000-token CLAUDE.md adds $0.009 per Sonnet message and $0.045 per Opus message. Over 200 messages/day, that is $1.80-$9.00/day just for the system prompt. Trim your CLAUDE.md aggressively.

Tool definitions: Each enabled tool (Bash, Read, Write, Glob, Grep, etc.) adds 200-400 tokens to every request. Six default tools add ~1,500 tokens. See cost optimization strategies.

Retries and compaction: When Claude Code retries failed requests or runs /compact, those are additional API calls that count against your quota or token budget.

Extended thinking: Thinking tokens are billed at output rates. A 2,000-token thinking block on Opus costs $0.15 – as much as an entire regular Sonnet response.

Choosing the Right Model

Not every task needs Opus. Use the right model for each task type:

  • Haiku: Quick questions, simple file reads, syntax checks, formatting
  • Sonnet: Standard coding tasks, refactoring, bug fixes, test writing
  • Opus: Complex architecture decisions, multi-file refactors, debugging subtle issues

The Model Selector tool recommends the optimal model based on your task description.

Try It Yourself

Stop estimating costs in your head. The Cost Calculator takes your usage patterns (hours per day, task types, model preferences) and shows your exact monthly cost across all plans. Compare Pro vs Max vs API Direct side by side.

Try the Calculator –>

Common Questions

Does Claude Code cost extra on top of the subscription? No. Pro and Max subscriptions include Claude Code access at no additional charge. The subscription covers both the web interface (claude.ai) and the CLI tool (Claude Code). API Direct is the only option where you pay per token.
Can I switch between plans mid-month? Yes. Upgrading takes effect immediately and you pay the prorated difference. Downgrading takes effect at the next billing cycle -- you keep the higher tier until the current period ends.
Are extended thinking tokens billed separately? Extended thinking tokens are billed at the output token rate for your model. On Opus, thinking tokens cost $75 per million tokens -- the same rate as regular output tokens. On API Direct, this can add up quickly for complex reasoning tasks.
What happens when I hit rate limits on Pro? Claude Code queues your request and retries automatically with backoff. During peak hours, you may wait 30-60 seconds between messages. If limits are consistently disruptive, upgrade to Max for higher ceilings.

Estimate usage → Calculate your token consumption with our Token Estimator.