Claude Code cost per month
Short answer: $20/month covers a casual user, $100/month covers a serious user, $200/month covers a heavy user, and pay-per-token only makes sense if you're under $20 of usage or over $400. Here's how to figure out which bucket you're in.
The three ways to pay
| Plan | Price/month | What you get |
|---|---|---|
| Claude Pro | $20 | Claude Code included with the same usage limits as the Claude.ai chat (~45 messages every 5 hours of Sonnet, less for Opus) |
| Claude Max 5x | $100 | ~5× the Pro limits — covers a full workday of heavy Claude Code |
| Claude Max 20x | $200 | ~20× the Pro limits — for people running Claude Code continuously |
| API pay-per-token | Variable | Sonnet 4.6: $3/M input, $15/M output. Opus 4.7: $15/M input, $75/M output. Haiku 4.5: $1/M input, $5/M output |
What real usage looks like
Casual user — $20/month is plenty
You use Claude Code a few times a day. "Fix this bug." "Add a test for this function." "Explain this error." A handful of multi-turn conversations per day, each maybe 5–15 turns.
Verdict: Pro at $20/mo. You'll occasionally hit the 5-hour rate limit but recover within minutes.
Daily driver — $100/month makes sense
Claude Code is part of your loop most of the working day. You're delegating real tasks: "investigate this flake," "migrate this module to TypeScript," "rewrite this endpoint." Several long-running sessions per day. You hit Pro's limits frequently.
Verdict: Max 5x at $100/mo. Pays for itself by not blocking you when you're in flow.
Heavy user — $200/month is the right answer
You leave Claude Code running for hours. Multiple parallel sessions across repos. Background agents. Code review automation. You hit Max 5x limits in the afternoon.
Verdict: Max 20x at $200/mo. At this level the API would be $400–800/mo, so Max is genuinely cheaper.
Bursty / occasional — API pay-per-token
You don't use Claude Code daily but when you do, you use it heavily. Some months $50, some months $5. A subscription wastes money in the slow months.
Verdict: API key. Wire it into Claude Code via ANTHROPIC_API_KEY and pay only when you use it.
How to estimate your API cost
A real Claude Code conversation burns more tokens than you'd guess because it re-reads files on every tool call. Rough rules of thumb (Sonnet 4.6):
- Quick fix (5–10 turns, small repo): ~$0.10–0.30
- Multi-file feature (20–40 turns, mid-size repo): ~$1–4
- Investigation / refactor (60+ turns, large repo): ~$5–15
- "Just go fix the failing tests" (long-running, many tool calls): ~$10–30
20 quick fixes per day × 22 days = ~$66–130/month. Two big tasks per day × 22 days = ~$176/month and up.
Hidden costs and gotchas
Opus is 5× more expensive than Sonnet
Opus 4.7 input is $15/M, Sonnet 4.6 input is $3/M. Output is $75/M vs $15/M. Default to Sonnet. Switch to Opus only when Sonnet fails.
Tool calls multiply token count
Every Read, Bash, and Grep the agent runs gets re-sent in the next turn's context. A "small" task with 30 tool calls can cost more than a "big" task with 5.
Prompt caching cuts cost ~50%
Claude Code uses prompt caching automatically. Cached input tokens cost $0.30/M instead of $3/M. Long sessions on the same files benefit most.
Max plans don't include the full API limit
Max 20x is "20× Pro," not "unlimited." You can still hit limits during sustained continuous use. Anthropic has been clear that heavy abuse triggers throttling.
The honest pick
Trying it out: Claude Pro $20/mo. Or BYO API key with a $20 monthly cap.
Daily working dev: Claude Max 5x $100/mo. This is the sweet spot for most pros.
Power user: Claude Max 20x $200/mo. If you'd be over $300 on the API, Max is cheaper.
Bursty use: API pay-per-token. Set a hard budget cap.
Cost is only one variable. See Cursor vs Claude Code for whether Claude Code is even the right tool for your workflow.
See the all-in monthly cost of Claude Code + your IDE + your context layer in one place.
Estimate your full stack cost →