Claude Code cost calculator
Pro, Max, or API? Estimate what Claude Code will actually cost you.
Pick how you use Claude Code. Get an API-equivalent monthly estimate and a plan recommendation in 30 seconds.
Quick answer
- Use Claude Code casually a few times per week: Pro is probably enough.
- Daily coding, occasionally hitting limits: Pro first, Max 5x if you start hitting usage windows.
- Heavy agentic sessions, long context, lots of tool calls: Max 20x is often the right fit.
- Automation, CI, or background agents: API beats subscription regardless of cost.
- The calculator estimates API-equivalent usage. Subscription rate limits aren't dollar-billed, so your subscription cost is fixed at the plan price.
Daily app dev — debugging, refactoring, multi-step coding tasks across a few sessions per day. ~30 turns per session.
- Cacheable input (system + tools) $0
- Variable input (user + tool results) $0
- Output (incl. tool calls) $0
- Subtotal × retry $0
- Without caching $0
- With caching (current setting) $0
- Caching saves $0
Claude Code reuses system prompts and tool definitions, so prompt caching applies aggressively.
▸ Edit assumptions 10 inputs
Claude Code typically hits 75–90% on system + tools because they're stable across turns within the cache TTL.
This estimate is the API-equivalent value of your usage at Anthropic's published per-token rates. Subscription plans (Pro / Max) are rate-limited, not dollar-billed, so your subscription cost is fixed at the plan price. Real overage behavior, plan limits, and pricing can change — check anthropic.com/pricing.
Pro vs Max vs API
Same workflow, different shapes. Match the one that fits your usage pattern.
| Plan | $/mo | Best for | Watch out for |
|---|---|---|---|
| Claude Pro | $20 | Casual builders, weekend projects, light daily Claude Code use | May hit usage limits during long agentic sessions |
| Claude Max 5x | $100 | Daily developers who use Claude Code heavily | Fixed cost only makes sense if you use it enough |
| Claude Max 20x | $200 | Very heavy users, long sessions, frequent limit frustration | Expensive if usage is inconsistent |
| Anthropic API | usage | Automation, CI, background agents, predictable usage-based workflows | Can get expensive if loops run uncontrolled |
Plan prices verified 2026-05-04 against anthropic.com/pricing. Anthropic adjusts plan structure occasionally; confirm current rules before committing.
What actually drives the cost
Long context windows
Claude Code reads files into context as it works. The more it needs to hold in memory across turns, the more tokens each turn sends. A 50-turn debugging session against a 30k-token codebase quickly costs more than 50 small one-shot prompts.
Multi-turn agent loops
Claude Code rarely answers in one shot. A "fix this bug" task can be 30+ turns of read / edit / run / verify. Each turn re-sends most of the context. The calculator's "turns per session" input captures this.
Tool calls and tool results
Read, Write, Bash, Edit, Glob, Grep — every tool call returns content that lands back in context. A single Bash call returning 2,000 lines of test output is a 2,000-token addition. Tool calls compound across turns.
Retry cycles
Failed commands, bad patches, unexpected errors — Claude Code re-runs prior steps. The retry-rate input (default 10%) captures this overhead. For agentic / experimental work 15–25% is more realistic.
Prompt caching
The big lever in Claude Code's favor. The system prompt and tool definitions are stable, so Anthropic's prompt cache applies aggressively — cache reads are roughly 10× cheaper than fresh input. The "Caching impact" panel shows what your cost would be without it. Spoiler: a lot more.
Subscription limits aren't dollar-billed
Pro and Max plans cap usage by rate limits and 5-hour windows, not by dollars. This calculator estimates the API-equivalent value of your usage so you can pick the plan whose limits fit. Once you're on a plan, you don't pay overage in dollars — you just hit a limit.
Who fits which plan
Rough profiles. Your mileage varies; use the calculator to check.
- Weekend builder. A few short Claude Code sessions per week, mostly small file edits and explanations. Pro is almost always enough.
- Daily indie dev. 1–3 sessions per workday on their main project. Some days hit the 5-hour limit. Pro to Max 5x depending on session length.
- Heavy agentic user. Long debugging or refactor sessions across larger codebases, frequent tool calls, occasional limit frustration. Max 5x or Max 20x typically.
- Automation user. Background scripts, CI agents, scheduled coding tasks. API almost regardless of cost — subscription rate limits aren't designed for this shape.
- Account-rotating user. Some people split usage across several Pro accounts to multiply the rate-limit budget. The calculator output is per-account; if you rotate 3 Pro accounts, divide projected usage by 3 for any single account. Anthropic's terms govern this — confirm whether your usage is allowed.
Assumptions
- This calculator estimates API-equivalent usage, not direct subscription cost.
- Default model is Claude Sonnet 4.6, Claude Code's primary model. Switch to Opus 4.7 in advanced settings to see what heavy Opus use would cost.
- Default cache hit rate is 80%. Real Claude Code sessions with stable system prompts often hit 75–90%.
- Cache writes use Anthropic's 1-hour TTL rate where available, since Claude Code holds context for hours within a session.
- Retry rate defaults to 10% for daily use, 15% for heavy/agentic. Real workloads vary widely.
- Token counts per turn are estimates derived from typical Claude Code session shapes. Override any number in "Edit assumptions" if you have better data.
- Plan prices and rate-limit policies change. Last verified: 2026-05-04. Always confirm against anthropic.com/pricing and the Claude Code docs before relying on these figures for budget decisions.
FAQ
Is Claude Code included in Claude Pro? ＋
Yes. Claude Pro ($20/mo) includes Claude Code access at Pro rate limits. Light to moderate users typically stay within those limits. Heavier users hit the 5-hour usage windows and may need to upgrade to Max.
Is Claude Code cheaper than the Anthropic API? ＋
For interactive coding work, the subscription plans (Pro, Max) are usually cheaper than equivalent API usage because of how Claude Code uses long contexts and prompt caching. For background automation or CI workloads, the API is often better — predictable per-token billing, no rate-limit surprises.
When is Claude Max worth it? ＋
Once your usage routinely hits Pro's 5-hour usage limits, or when long agentic sessions feel cut short. Max 5x ($100/mo) gives roughly 5× Pro's limits; Max 20x ($200/mo) gives roughly 20×. The calculator above suggests a tier based on your API-equivalent usage.
Why does Claude Code usage feel unpredictable? ＋
Three reasons: (1) context grows turn-by-turn as more files get read, (2) tool calls and tool results add tokens, (3) retries from failed commands re-run earlier turns. A single coding task can quietly become 50–100 turns. The calculator models all three.
Should I use Claude Code or Cursor? ＋
Different shapes. Claude Code is terminal-first, agent-first, and uses subscription rate limits. Cursor is editor-first with mixed pricing (subscription + bring-your-own-key options). If your workflow is multi-step terminal-based agentic coding, Claude Code's subscription often wins on cost. If you want IDE-integrated autocomplete, Cursor is better.
Should I use Claude Code subscription or API? ＋
Subscription if you're an indie developer with daily interactive coding sessions — fixed monthly cost, no surprises. API if you're running automation, CI agents, or workloads where you need precise per-token cost control.
Can this calculator predict my exact bill? ＋
No. It estimates the API-equivalent value of your usage. Subscription plans are rate-limited, not dollar-billed, so your subscription cost stays fixed at the plan price. The calculator helps you decide which plan best fits your usage level.
Does prompt caching reduce Claude Code cost? ＋
Yes — significantly. Claude Code reuses system prompts and tool definitions across turns, so prompt caching applies aggressively. Cache reads are roughly 10× cheaper than fresh input. The "with vs without caching" panel above shows the impact.
Disclaimer. This page is an estimator, not a bill. Actual Claude Code limits, plan structure, and pricing may differ from what's modeled here. Anthropic adjusts plan rules from time to time. Use this calculator to make a directional decision, then confirm against official sources.
For a more general workload simulator (any agent, any model), see the general agent cost calculator. For caching and batch decisions, see caching ROI and batch vs real-time.