Claude Code is included in every paid Claude plan. On monthly billing, individuals pay $20 for Pro, $100 for Max or $200 for Ultra. For teams, Claude Code pricing starts at $20 per engineer per month on annual billing, rising to $100 for a premium seat. Enterprise charges $20 per seat plus usage at API rates. Anthropic reports average usage costs of $150 to $250 per developer per month across enterprise deployments, or around $13 per active day. For 90% of users, the daily figure stays below $30.
At 100 engineers, standard Team seats cost $2,000 a month. Give everyone premium and the bill becomes $10,000, before any extra usage. You probably have people who open Claude Code a few times a week alongside people who use agentic coding all day. They do not all need the same allowance. To budget sensibly, you need to know who reaches their limits, what it costs to keep them working and how much of that work gets merged.
How much does Claude Code cost?#
If you already pay for Claude, you do not need to buy Claude Code separately. The Claude Code plans below include access to the terminal tool alongside chat and Claude Desktop. All three use the same allowance. The advertised Claude Code price therefore covers your use across all three, within the plan's limits.
| Plan | Price | Claude Code and usage |
|---|---|---|
| Free | $0. | Claude Code is not included. |
| Pro | $20 a month billed monthly, or $17 a month billed annually ($200 up front). | Claude Code is included. |
| Max | $100 a month, billed monthly. | You get 5x Pro usage per 5-hour session and per week. |
| Ultra | $200 a month, billed monthly. | You get 10x Pro usage per week, and no 5-hour limit for verified accounts. |
| Team, standard seat | $25 per seat a month billed monthly, or $20 billed annually. | Claude Code is included in every seat, for teams of 2 to 150. |
| Team, premium seat | $125 per seat a month billed monthly, or $100 billed annually. | Each seat gets 5x the usage of a standard seat, and seat types can be mixed. |
| Enterprise | $20 per seat a month billed annually, with a 20-seat minimum. | Usage is billed at API rates on top of the seat price, and Anthropic does not publish discounts. |
When comparing Anthropic Claude Code pricing with other guides, check which plans they describe. Some still list "Max 20x", while Anthropic now lists Pro, Max and Ultra. Others say standard Team seats exclude Claude Code, although Anthropic's help centre confirms that it is included in every Team seat. Premium buys more usage. There is a separate restriction on Team's pooled API credits: the plan includes up to $500 a month, but you cannot spend those credits in Claude Code or the Claude apps.
How do Claude Code usage limits work?#
An engineer can reach their Claude Code limit even if they have barely used the terminal that day. Work in claude.ai and Claude Desktop counts towards the same allowance. Most subscription plans have a five-hour session limit, measured from the first message, as well as weekly limits that reset at a fixed time assigned to the account. Verified Ultra accounts have no five-hour limit. Moving from chat to the terminal does not start a fresh allowance.
When Claude stops you, read which limit you have reached. An Opus-specific limit still lets you switch to another model with /model. A session or weekly limit that covers all models leaves you waiting for the reset unless you upgrade or pay for more usage. On Pro, Max, Ultra and Team, you can enable usage credits to keep working at standard API rates. That is where a flat monthly subscription can start to accumulate additional charges.
You cannot reliably convert the advertised allowance into a fixed number of tokens. Anthropic publishes multiples of Pro usage, not token counts for each window. Figures quoted by other blogs are therefore a poor basis for deciding how much work a seat will cover. Anthropic also reserves the option to introduce weekly or monthly caps and model-specific limits, so check the terms again before committing to a rollout.
What does Claude Code cost on the API?#
A flat Claude Code subscription cost pays for an allowance. With the API, each request adds to the bill according to the model and the number of tokens it uses. This applies when you use the Claude Console, a cloud provider or usage-based Enterprise billing; Enterprise also has a seat fee. Claude Code API pricing uses the standard model rates below, quoted per million tokens.
| Model | Price per million tokens |
|---|---|
| Opus 5.5 | $4 for input and $20 for output, with cache reads at $0.20. |
| Sonnet 5.5 | $2 for input and $10 for output, with cache reads at $0.10. |
| Haiku 5.5 | $0.10 for input and $0.50 for output on prompts up to 100,000 tokens, with cache reads at $0.01. |
At these rates, a million output tokens cost $20 on Opus and $10 on Sonnet. That difference adds up when agents keep working through a task, which is why Anthropic recommends Sonnet for most coding and Haiku for simple subagent work. Cache writes are priced separately: a 5-minute cache costs 1.25x the input rate, and a 1-hour cache costs 2x. Comparing the Claude Code token cost with older models also requires care. Claude 4.7 and later use a tokenizer that produces about 30% more tokens for the same text, so the rate alone understates the difference in cost.
What does Claude Code cost an engineering team?#
For 20 engineers, standard Team seats come to $400 a month on annual billing. At the same team size, Enterprise comes to an estimated $3,400 to $5,400 when we add Anthropic's reported average usage. Our model uses the published seat prices for Team and, for Enterprise, $150 to $250 in monthly usage plus the $20 seat fee. That makes Enterprise $170 to $270 per engineer. The larger-team calculations follow the same assumptions.
| Engineers | Team, all standard | Team, 1 in 4 premium | Team, all premium | Enterprise with Anthropic's average usage |
|---|---|---|---|---|
| 20 | $400 | $800 | $2,000 | $3,400 to $5,400 |
| 50 | $1,000 | $2,000 | $5,000 | $8,500 to $13,500 |
| 100 | $2,000 | $4,000 | $10,000 | $17,000 to $27,000 |
| 200 | Not available, because Team covers 2 to 150 seats. | Not available. | Not available. | $34,000 to $54,000 |
Monthly cost with annual billing, in US dollars.
Read the Team columns as the subscription bill, before extra usage. An engineer who reaches their allowance and carries on with usage credits will add charges at API rates. We have not estimated those charges here. The Enterprise column includes usage, but it uses Anthropic's averages across enterprise deployments. Your pilot needs to establish how close your own engineers come to those figures.
The $100 premium Team seat is cheaper than the reported $150 to $250 average monthly usage bill, provided the engineer can work within its limits. Enterprise removes per-seat usage limits. That can help when agents run in parallel or continue working unattended, but it also means the bill keeps growing as they work. Set spend caps before enabling that usage.
Claude Code team pricing#
Start by separating occasional users from engineers who use Claude Code throughout the day. Team lets you buy standard seats for the former and premium seats for the latter, at $20 and $100 a month respectively on annual billing. For 100 engineers, putting one in four on premium costs $4,000 a month. That is twice the all-standard bill, but well under half the $10,000 cost of premium for everyone. Team supports up to 150 seats; above that, you need Enterprise.
Claude Code enterprise pricing#
Budget Enterprise at the published $20 seat fee plus usage unless you have a different quote from Anthropic. The company does not publish discount rates, so there is no discount assumed in our figures. Every Enterprise seat on new and self-serve plans includes Claude Code. Admins can cap spend for individual users and the organisation, which matters when there is no per-seat usage allowance to stop further work.
Figure 1. Team figures are seat fees only and exclude usage credits for engineers who pass their limits. Enterprise is $20 per seat plus Anthropic's reported average usage of $150 to $250 per developer a month. Team plans cover up to 150 seats. Sources: claude.com/pricing and the Claude Code costs documentation, November 2026.
Enjoying it so far?Get more like it in our AI Native newsletter.
Read the newsletterWhy does Claude Code seat pricing tell you so little about cost per change?#
Suppose an agent writes a change, fails a build and tries again. You pay for both attempts. If review sends it back for another revision, that costs tokens too, even if the change never merges. A monthly seat price tells you little about how much of this repeated work is happening. We measure cost per merged change so we can see how much AI spend it takes to get usable code through review, including the money spent on unsuccessful attempts.
In October 2026, we tracked a weekend of work in Lore, re:cinq's internal software factory platform. Its autonomous coding loop produced pull requests for 54 tickets, and 23 were merged. A merged ticket typically cost $21 in AI spend, with individual costs ranging from $6 to $63. Lore uses Claude Sonnet and Gemini models, so these figures are not a Claude Code subscription bill. They show what it cost to get changes merged in a coding loop that uses several models.
The most useful finding was where the repeated work came from. Failed builds and review comments led to $709 in extra rounds. Of that, $368 concerned lint and spec-link rules the agent was not allowed to check before pushing. It had to wait for the failure before it could respond. Running those checks first would have saved about 170 rounds and $302, without changing the subscription or model.
The tool and model still affect the result. In July 2026, Veli-Matti Vanamo at Insight APAC measured what it cost to merge one feature using 23 combinations of coding tool and model. The cost ranged from $2.59 to $33.38 for that one feature. Even among the Claude Code combinations, it ranged from $9.98 to $33.38.
Before moving everyone to a cheaper plan, look at what agents are being asked to redo. In Lore, we could identify specific checks behind the extra spend. Checks that run before the agent pushes, such as lint, type checks and spec-link checks, let it catch problems before a failed build sends it through another round. Tracking those failures alongside cost per merged change gives you something concrete to fix.
What makes a Claude Code bill go up?#
Running a task on Opus costs twice as much per token as running it on Sonnet at the rates above. If Sonnet can do the work, the extra cost needs a reason. Running several agents adds another source of spend: each teammate maintains its own context window. Anthropic reports about 7x the token usage of a standard session when agent teammates run in plan mode. A team of agents can therefore consume an allowance much faster than one engineer working with a single session.
Long sessions can also make a small request more expensive than it looks. The model receives earlier conversation as context, including work you may already have finished. Use /clear when you move to an unrelated task so that history does not keep travelling with new requests. The same principle applies to instructions and tools loaded into a session. CLAUDE.md loads at the start of every session, which is why Anthropic recommends fewer than 200 lines. Configured MCP servers can add overhead even when no one uses them.
Check the reasoning effort setting as well. Extended thinking is billed as output tokens, so high effort on a routine task means paying for reasoning the task may not need. The /effort command lets you lower that setting.
How do you control Claude Code costs for a team?#
An engineer checking a session and a manager checking the team bill need different views. In Claude Code, /usage gives a session estimate. For the API bill, use the Usage page in the Claude Console as the authoritative record. OpenTelemetry export lets you bring per-user token and cost metrics into your own monitoring in near real time, and works across setups.
Put limits in place before enabling extra usage. A Team seat stops at its included allowance, but usage credits allow the engineer to continue at additional cost. Team and Enterprise admins can set spend limits for the organisation, groups or individual members; API organisations can also cap workspace spend in the Console. Rate limits need separate attention as an API deployment grows. Anthropic recommends 200,000 to 300,000 tokens per minute per user for teams of 1 to 5, falling to 15,000 to 20,000 for teams of 100 to 500. A larger team is less likely to have every user active at once.
Give engineers a starting setup they can use without making a cost decision on every task. Anthropic recommends Sonnet for everyday coding and Haiku for simple subagents. Keep CLAUDE.md short and put detailed workflow instructions in skills. Clear sessions between unrelated tasks, disable unused MCP servers, and keep agent teams small enough for the work. Shut those teams down when they finish.
After 30 days, review each engineer's spend against the cap you set before rollout, and measure cost per merged change. Someone repeatedly reaching a limit may need a premium seat or more usage. Repeated build failures may call for better checks before you spend more. Reviewing both helps you decide where to change the plan and where to change how the work is done. If you are rolling Claude Code out across a team, re:cinq offers agentic coding training.
Frequently asked questions#
Is the $20 Claude Code plan worth it? For occasional coding work, Pro gives you access for $20 a month. It has the smallest paid allowance, so an engineer using it heavily every day may reach the five-hour session limit. Max and Ultra provide more usage for people who need it.
Is Claude Code completely free? No. The Free plan excludes Claude Code. The lowest subscription price is Pro at $17 a month on annual billing, or $20 when paid monthly.
Does it cost money to code with Claude? Yes. You can pay through a Claude subscription or use pay-as-you-go API access. API charges depend on your model and the number of tokens used.
Is Claude Code included in the Team plan? Yes, including standard seats. Premium seats include the same tool with 5x the standard usage allowance.
How much does Claude Code cost per month? On monthly individual plans, the price is $20 to $200. Team seats range from $20 to $125 per engineer depending on the seat and billing period. For usage-based enterprise deployments, Anthropic reports average usage of $150 to $250 per developer per month, with the seat fee charged separately.
Does Claude Code charge per token? API usage and usage credits are charged per token. A subscription covers usage within a set allowance for a flat fee; enabling usage credits adds per-token charges after that allowance runs out.
Sources and further reading#
- Claude plans and pricing, Anthropic: individual, Team and Enterprise prices and usage multipliers.
- Manage costs effectively, Claude Code documentation: average spend per developer, cost controls and rate-limit guidance.
- Pricing, Claude API documentation: per-token model rates, caching and the tokenizer note.
- Use Claude Code with your Team or Enterprise plan, Anthropic help centre: which seats include Claude Code.
- How do usage and length limits work, Anthropic help centre: session and weekly limits and usage credits.
- Lore implementation loop, weekend of 23 October 2026, re:cinq: cost per merged ticket in our autonomous coding loop.
- Cost to a merged feature, Insight APAC: one feature merged across 23 model and tool combinations.
Related articles#
- What is agentic coding?: the practice Claude Code is built for, and the evidence behind it.
- FinOps for AI: what AI spend should be measured against.
- AI Cost Optimization for Coding Agents: tracking coding-agent costs by issue and repository.
- Why Your 200K Context Window Is Smaller Than You Think: how long sessions fill the context window.







