Option 1: a Claude subscription
| Plan | Monthly | Claude Code allowance |
| Pro | $20 ($17 on annual) | ~10–40 prompts / 5 h · ~40–80 Sonnet hrs/wk · no Opus |
| Max 5x | $100 | ~50–200 prompts / 5 h · ~140–280 Sonnet + 15–35 Opus hrs/wk |
| Max 20x | $200 | ~200–800 prompts / 5 h · ~240–480 Sonnet + 24–40 Opus hrs/wk |
| Team Standard seat | $25 ($20 annual) | More than Pro; Claude Code included |
| Team Premium seat | $125 ($100 annual) | 5× Standard; admin spend controls |
The subscription is a flat price for a metered allowance. You never see a token bill; you see a usage bar (/usage) and, when it fills, a pause until the 5-hour or weekly window resets. If you'd rather not pause, usage credits let extra usage bill at standard API rates up to a monthly limit you set.
Option 2: the API, pay per token
Sign in with a Claude Console account instead of a claude.ai subscription and Claude Code bills every request to your API balance at list price. Current per-million-token prices:
| Model | Input | Output | Cache read |
| Claude Haiku 4.5 | $1 | $5 | $0.10 |
| Claude Sonnet 5 | $2 | $10 | $0.20 |
| Claude Opus 5 | $5 | $25 | $0.50 |
| Claude Fable 5.1 | $10 | $50 | $0.25 |
Cache reads matter more than the headline prices. Claude Code resends your whole conversation on every request, but with prompt caching the unchanged prefix is read at a tenth of the input price (a fortieth for Fable). In a healthy session, 90% or more of input tokens come from cache; /usage shows the share.
Anthropic's published benchmark across enterprise deployments: about $13 per developer per active day, $150–250 per developer per month, with 90% of users under $30 per active day. Two habits explain most of the users above that: sessions that were never cleared, and Opus left as the default model.
What an hour actually costs
Per-token prices are hard to feel. Converting Anthropic's benchmark and the per-token math into hours of active Claude Code on Sonnet, with caching working:
| Session type | API-equivalent per hour (Sonnet) | Opus (~2.5×) | Fable (~5×) |
| Light: short tasks, small files, cleared often | ~$1.50 | ~$4 | ~$7.50 |
| Moderate: typical feature work | ~$3.50 | ~$9 | ~$17.50 |
| Heavy: large repo, long sessions, subagents | ~$8+ | ~$20+ | ~$40+ |
Those are the assumptions behind the API cost calculator, which lets you change them. A moderate three-hour day on Sonnet is roughly $10, which is exactly where Anthropic's $13 average lands once you add a bit of Opus.
The break-even
Compare the plan price to the API-equivalent of your usage:
- Pro, $20: beaten by the API only if you use Claude Code fewer than about 6 moderate Sonnet hours a month. Two afternoons.
- Max 5x, $100: beaten by the API below about 28 moderate Sonnet hours a month, roughly 1.5 hours per workday. Above that, the subscription is cheaper, and by a full-time user's 80–120 hours it's a fraction of the API bill.
- Max 20x, $200: for anyone using it enough to need it, the API-equivalent is $400–800+ a month.
So the rule is simple: occasional use, API; daily use, subscription. The exception is anything unattended: CI jobs, scheduled agents, a bot on a server. Subscriptions are for a person at a keyboard; automation runs on the API, and the agent running-cost calculator covers that case.
What a team should budget
Anthropic's own guidance to admins: budget more for a coding seat than a chat seat, because each Claude Code turn carries file contents, tool calls and multi-step reasoning, and one debugging session can consume more than a day of chat. On the API, $150–250 per developer per month is the planning figure; set a workspace spend limit in the Console so a runaway session can't exceed it. On Teams or Enterprise, the seat allowance is the ceiling by default, and usage credits with per-member limits let heavy users continue. Either way, run a pilot group with /usage and the spend report before rolling out, because the spread between a light and a heavy developer is 5–10×.
Hidden costs and non-costs
- Background usage (session summaries for
--resume, status checks) is real but tiny, typically under $0.04 per session. - Idle sessions aren't free if they have scheduled tasks, agent teammates or cross-session messaging active; each fires with your full context.
- Extended thinking tokens bill as output. Lowering the effort level with
/effortcuts the bill on simple tasks. - Data-residency API requests bill at 1.1× list price.
- Cursor, Copilot and Codex price the same tokens differently: included-usage wallets or message counts rather than hours. The Claude Code vs Cursor guide and the tools comparison line them up.
Calculate my API cost with caching →
Frequently asked questions
How much does Claude Code cost?
Free to install. Usage is included in Pro ($20), Max ($100/$200) and Team seats, or billed per token on the API (Anthropic's average: ~$13 per developer per active day).
Is Claude Code free with Pro?
Yes, with Sonnet-class models and roughly 40–80 Sonnet hours a week. No Opus in Claude Code on Pro.
How much does an hour cost on the API?
About $1.50–8 on Sonnet depending on context size, 2.5× that on Opus, 5× on Fable.
Is the API cheaper than Max?
Only below roughly 28 moderate Sonnet hours a month. Daily users save with the subscription.
Can I use Claude Code in CI on a subscription?
Subscriptions are for interactive use; unattended automation runs on the API.
What should a company budget per developer?
$150–250/month on the API, or a $125 Premium Team seat; pilot first, the per-developer spread is large.
Sources
- Claude Code docs: Manage costs effectively ($13/day and $150–250/month benchmarks, background usage, team guidance).
- Anthropic: Claude plans and pricing.
- Claude Platform: model pricing (per-million-token rates, cache and batch discounts).
- Anthropic support: Using Claude Code with your Pro or Max plan.