Claude Code cost: plans, API rates and what a session really costs

The short answer

Claude Code is paid per seat ($20 a month on Pro, $100 on Max, $200 on Ultra) or per token through the API. Anthropic reports an average of about $13 per developer per active day at API rates. On the API, our reference session costs $1.20 on Sonnet 5.5 and $2.40 on Opus 5.5.

Claude Code is billed in one of two ways. On a subscription you pay a monthly seat and work inside a usage allowance. On the API you pay for every token, and the bill follows the model, the size of the context and how well the prompt cache holds. This guide gives the public prices, the average Anthropic publishes for a developer, and the cost of one session broken down by token class. Prices are US list prices in dollars, read on Anthropic's pricing pages on November 13, 2026.

What each plan costs

PlanMonthly priceUsage
Pro$20 ($17 billed annually)the first plan that includes Claude Code
Max$100five times Pro, per five-hour window and per week, plus $100 of API credits a month
Ultra$200no five-hour limit for verified accounts, ten times Pro per week, plus $200 of API credits a month
Team, standard seat$25 per seat ($20 billed annually)more usage than Pro, 2 to 150 seats, up to $500 of pooled API credits a month
Team, premium seat$125 per seat ($100 billed annually)five times a standard seat
Enterprise$20 per seat, billed annuallyusage is billed on top, at API rates

The free plan does not include Claude Code. Anthropic states the allowances as multiples of Pro and gives no number of messages or tokens. Its pricing page adds that how fast you reach a limit depends on the model, the length of the conversation and the size of the files.

A seat is not always a fixed cost. When someone reaches the limit, they can turn on usage credits. Requests beyond the allowance are then billed at standard API rates, separately from the subscription, up to a monthly spend limit if one is set. The monthly API credits included in Max and Ultra do not apply to them.

What the API costs

Anthropic's list prices per million tokens:

ModelInputOutputCache write, 5 minutesCache write, 1 hourCache read
Claude Fable 5.2$8$40$10$16$0.20
Claude Opus 5.5$4$20$5$8$0.20
Claude Sonnet 5.5$2$10$2.50$4$0.10
Claude Haiku 5.5$0.10$0.50$0.125$0.20$0.01

Output costs 5 times input across the range. Writing to the cache costs 1.25 times input for a five-minute lifetime and 2 times for one hour. Haiku 5.5 has a second tier: a request whose prompt is over 100,000 tokens, cache included, is billed at $0.50 for input and $2.50 for output. Per-model energy and CO₂e figures sit next to these prices on the model sheets.

What Anthropic reports per developer

In its Claude Code documentation, Anthropic gives an order of magnitude for enterprise deployments at API rates: about $13 per developer per active day on average, $150 to $250 per developer per month, and under $30 per active day for 90% of users. It is an average over many teams. The same page says costs vary widely with the model, the size of the codebase and habits such as running several instances at once.

What one session costs

A Claude Code session sends the whole conversation with every request, so most of the tokens you pay for are context being read again. Take our reference session on Sonnet 5.5, a TokenClimate assumption: 50,000 fresh input tokens, 200,000 tokens written to the cache, 3,000,000 read from it, and 30,000 generated.

At the five-minute cache rate that session costs $1.20. Cache reads are 91% of the tokens and 25% of the cost. Output is 0.9% of the tokens and 25% of the cost. Context as a whole, fresh input and cache together, is 75% of the bill.

The same volumes cost $2.40 on Opus 5.5 and $4.20 on Fable 5.2. On Haiku 5.5 they cost $0.075 if every request stays under the 100,000-token threshold, and $0.375 above it. With a one-hour cache, the Sonnet 5.5 session comes to $1.50. The chapter why long context costs so much of our course explains why each turn is billed again for everything that came before it.

What moves the bill

Three things account for most of the gap between two developers.

The default model comes first. Anthropic's own advice is Sonnet for most coding tasks, Opus for complex architecture decisions and multi-step reasoning, and Haiku for simple subagent tasks.

Session length is next. A one-line question asked in a session that has been open all day is still billed for the whole conversation. Anthropic names clearing the context between unrelated tasks as one of the two habits with the most effect, next to matching the model to the job.

The cache is the third. The first message after a break longer than the cache lifetime reprocesses the full context. On a subscription, within the plan's included usage, Claude Code requests a one-hour lifetime for the main conversation. With an API key or on usage credits the default is five minutes. Switching model mid-session also starts a new cache, since each model has its own.

Time spent in the tool is not on that list. A long stretch of thinking over a small context costs less than a string of short exchanges over a large repository.

Subscription or API

For each person, compare two numbers. The first is what you actually pay: the seat, plus anything billed per token on top. The second is what the same tokens would have cost at API list prices. When the second is well above the price of the seat, the seat is good value. When it stays below, a smaller plan is enough. Cutting a seat that pays for itself moves the same usage to the API at full price.

That comparison needs token counts per person and per model, which a seat invoice does not show. Anthropic notes that usage inside the seat allowance is not metered in dollars. The calculator prices a volume of tokens at list rates and puts the CO₂e next to it. To run the comparison on your team's real usage, see how we work with a team.

FAQ

Is Claude Code free?

No. The free Claude plan does not include Claude Code. The cheapest way in is the Pro plan at $20 a month, or an API key billed per token.

How much does Claude Code cost per month for a developer?

On a subscription, the seat: $20 on Pro, $100 on Max, $200 on Ultra. On the API, Anthropic reports $150 to $250 per developer per month on average across enterprise deployments. Your own figure depends on the model and on context size.

Is the Max plan cheaper than the API?

It depends on the volume. Max costs $100 a month for five times the usage of Pro. If the tokens a person uses would cost more than that at API rates, the plan is the cheaper option. If they would cost less, it is not.

What happens when I hit the usage limit?

Claude Code shows when the limit resets. You can wait, move to a higher plan, or turn on usage credits, which are billed at standard API rates on top of the subscription.

Does prompt caching reduce the cost?

Yes, when the cached context is read many times. On Sonnet 5.5 a cache read costs $0.10 per million tokens against $2 for fresh input, and a five-minute cache write costs $2.50. A prefix that is written and never read again costs more than sending it uncached.

Sources

Guide written by Gaëtan Wittebolle for TokenClimate.