Claude Code Cost: What You'll Actually Pay (Subscription vs API)

Quick answer

Claude Code cost comes down to how you access it: on a Claude subscription (Pro at $20/month or Max at $100–$200/month) usage is bundled in, so you pay nothing extra per session. On the API you pay per token — roughly $3–$15 per million tokens on Sonnet — which for most developers works out to around $6–$13 per active day.** The subscription is the predictable option; the API scales with exactly how much you run. We run this site's tooling with Claude Code every day, so we've watched the meter closely. Below is the honest breakdown of your total Claude Code cost on each route, real per-day numbers, and the two commands and habits that keep the bill from creeping up. If you only want the "is it free" question answered, start with our guide on whether [Claude Code is free](/is-claude-code-free); this piece is about what it actually costs once you're past the free tier. ## The two ways Claude Code cost works There is no separate "Claude Code plan." You pay for Claude Code one of two ways, and picking the right one is the single biggest lever on your bill. | Route | How you're billed | Best for | |---|---|---| | **Claude subscription** (Pro / Max) | Flat monthly fee; Claude Code usage included within your plan limits | Individuals, predictable budgets, daily interactive coding | | **API key** (pay-as-you-go) | Per token, per model — no monthly minimum | Teams, automation, CI, spiky or high-volume usage | On a subscription, the `/cost` figure Claude Code shows is irrelevant to billing — you've already paid a flat fee and usage draws from your plan's limits. On an API key, every token is metered. That difference is why the same workload can feel free one month and cost $200 the next, purely based on which door you walked through. ## Claude Code subscription cost: Pro, Max, and what's included The subscription route makes your Claude Code cost a fixed line item. You install Claude Code, sign in with your Claude account, and coding usage counts against your plan's rolling limits rather than a token meter. | Plan | Price / month | Claude Code access | Roughly who it's for | |---|---|---|---| | Free | $0 | Not included — Claude Code needs a paid plan or API credits | Trying the chat app only | | Pro | $20 | Included, tighter usage window (Sonnet-class models) | Solo devs, side projects | | Max 5× | $100 | Included, ~5× the Pro usage, Opus access | Heavy daily users | | Max 20× | $200 | Included, ~20× the Pro usage | All-day agentic coding | Pro is the cheapest way to use Claude Code seriously at $20/month. The catch is the usage window: run a few long agentic sessions and you'll hit Pro's cap, at which point Max 5× ($100) or Max 20× ($200) buys you more headroom rather than a better model. For the full plan-by-plan math on the consumer side, see our [how much Claude costs](/how-much-does-claude-cost) guide — the subscription tiers there are the same ones that unlock Claude Code.  ## Claude Code API cost: pay per token If you use an API key instead of a subscription, your Claude Code cost is pure usage — input tokens plus output tokens, priced per model. There's no monthly minimum and no usage cap, which is powerful and dangerous in equal measure. These are the current per-million-token rates you'll see reflected in your [API token pricing](https://platform.claude.com/docs/en/about-claude/pricing): | Model | Input / MTok | Output / MTok | When Claude Code uses it | |---|---|---|---| | Claude Opus 4.8 | $5 | $25 | Hardest refactors, deep reasoning | | Claude Sonnet 5 | $3 ($2 intro) | $15 ($10 intro) | The everyday default | | Claude Haiku 4.5 | $1 | $5 | Fast, cheap sub-tasks | Sonnet 5's introductory pricing ($2 input / $10 output per million tokens) runs through August 31, which makes the default model noticeably cheaper right now. Two levers cut the cost of Claude Code API usage further: **prompt caching** reuses the big unchanging context (your files, your `CLAUDE.md`) at about one-tenth the input price, and the model choice itself — dropping from Opus to Sonnet is a ~40% cost cut for near-identical everyday coding, which we break down in [Claude Sonnet vs Opus](/claude-sonnet-vs-opus). A worked example: a focused Sonnet session that reads a medium codebase and writes a feature might process ~2M input tokens (mostly cache reads) and ~250K output tokens. Uncached that's roughly $6–$10; with caching doing its job, closer to $2–$4. Multiply by how many of those you run in a day and you have your real API bill. ## What a typical Claude Code cost looks like per day Anthropic's own usage data is the most honest yardstick here. Across enterprise deployments, the average Claude Code cost lands around **$13 per developer per active day**, and **90% of users stay under $30 per active day**. Monthly, that shakes out to roughly **$150–$250 per developer** for heavy API users. | Usage profile | Rough daily API cost | Monthly ballpark | |---|---|---| | Light (a few tasks, Sonnet, cached) | $2–$6 | $40–$120 | | Typical developer (active day) | ~$13 | $150–$250 | | Heavy agentic (long Opus sessions) | $25–$30+ | $300+ | Read those numbers against the subscription tiers and the decision gets easy for most individuals: if your API spend would clear ~$20/month, Pro is cheaper; if it would clear ~$100/month, Max 5× is. The API only wins on raw cost when your usage is low and sporadic, or when you need it for automation a subscription can't cover. ## How to track and cut your Claude Code cost You can't manage a Claude Code cost you can't see. Claude Code ships with the tools to watch it in real time — [Anthropic documents the full set](https://code.claude.com/docs/en/costs), but these are the ones that matter day to day. - **Run `/cost`** — on an API key, `/cost` prints your running spend for the current session. (On a subscription it's informational only, since you've prepaid.) Treat a scary number as a signal, not a surprise. - **Clear stale sessions.** Unexpectedly high API bills almost always trace back to one very long session that was never cleared — the whole history gets re-sent and re-priced every turn. Start fresh when you switch tasks. - **Default to Sonnet, escalate to Opus per task.** Most work doesn't need the flagship. Learn to [change the model mid-session](/claude-code-plan-mode) and reserve Opus for genuinely hard problems. - **Let caching work.** Keep a stable `CLAUDE.md` and project context so the expensive input tokens get cached at ~10% price instead of paid in full each turn.  One caveat worth knowing: the `/cost` and SDK cost figures are **client-side estimates** computed from a bundled price table, not authoritative billing. For the real number, check the Usage page in your Claude Console. The estimate is close enough for steering; the Console is the source of truth. ## Subscription vs API: which gives the lowest Claude Code cost Here's the decision in one line: **if you code interactively most days, buy a subscription; if you automate or run spiky, high-volume workloads, use the API.