Anthropic's own figure for Claude Code is about $13 per developer per active day and $150 to $250 per developer per month, across enterprise deployments. Its Claude Code cost page adds that 90% of users stay below $30 on an active day. Those are usage costs, priced in tokens. What lands on your invoice also depends on how you buy it: seats with usage included, a seat plus usage at API rates, or every token billed to your account.
This post is for a CTO who has to put one line in a budget and defend it. It covers Anthropic's numbers, how the way you buy it changes the bill, a worked budget for 20 developers, and when to re-forecast. Every Claude price below is Anthropic's, read from its public pages on the day this post was published. Prices change, so check them again before you sign.
The Gnar Company is a Claude Select Services Partner in Anthropic's Claude Partner Network. For teams that want Claude working inside their own systems, Gnar does enterprise AI integration, starting with Gnar Discovery and Design: two weeks, $15,000, and you walk out with a build plan that has a fixed price on it.
What does Claude Code cost per developer?#
Anthropic's cost page gives three numbers: about $13 per developer per active day, $150 to $250 per developer per month, and under $30 per active day for 90% of users. They come from enterprise deployments and describe usage priced in tokens. A seat fee is separate.
| Measure | Anthropic's figure |
|---|---|
| Average per developer per active day | About $13 |
| Average per developer per month | $150 to $250 |
| 90% of users, per active day | Below $30 |
Three details matter before you put these numbers in a spreadsheet.
First, Anthropic's cost page does not define an active day. Read it as a day the developer used Claude Code. At the $13 average, a developer who uses it on 15 days in a month costs about $195. One who uses it on 20 days costs about $260, which is past the top of the monthly range. Anthropic does not say how many active days sit behind its monthly range. Treat the daily and monthly figures as two separate reference points.
Second, Anthropic says per developer cost varies widely with the model, the size of the codebase, and habits such as running several instances at once or automation. The average is a starting point. Your team's forecast comes from your pilot.
Third, Anthropic's own advice is to start with a small pilot group and use its tracking tools to set a baseline before a wider rollout. Anthropic's numbers set the order of magnitude, and your pilot sets the figure you budget.
Which way you buy it changes the bill#
Claude Code comes with every paid Claude plan, and you can also run it through the Console or a cloud provider. The bill takes three shapes. Team seats include usage up to an allowance. Usage-based Enterprise charges a seat plus usage at API rates. The Claude Console or a cloud provider bills every token to your organization.
Here is what Anthropic's pricing page lists for companies:
| Route | What you pay | What usage costs | Where you cap it |
|---|---|---|---|
| Team, Standard seat | $20 a seat a month billed annually, $25 billed monthly | Included up to the seat allowance | The allowance, then usage credits with an organization spend limit |
| Team, Premium seat | $100 a seat a month billed annually, $125 billed monthly | Included, at 5x the Standard allowance | Same as Standard |
| Enterprise usage-based plans | $20 a seat a month, billed annually, 20 seat minimum | Billed at API rates | Spend limits by organization, group and member |
| Console or cloud provider | No seat | Console: every token at Anthropic's API rates. Amazon Bedrock, Google Cloud and Microsoft Foundry: billed per token to your cloud account | Workspace limits or your cloud's budget controls |
Team is sold for teams of 2 to 150. On a Team seat, Claude Code shares one allowance with Claude chat. The allowance resets on a rolling five hour window and on a weekly window. Usage inside it is not metered in dollars. When a developer runs out, they wait for the reset, or an admin turns on usage credits. Anthropic bills usage credits at standard API rates.
On a usage-based Enterprise plan, including self serve Enterprise, Anthropic's support article says there are no per seat usage limits. On these plans the $20 seat fee covers access, and usage is billed separately at API rates. That is why Anthropic's per developer numbers matter most to Enterprise and API buyers. On those routes, usage on the scale of $150 to $250 a month is billed on top of any seat fee.
The API rates set the price of every token. Anthropic's pricing page lists these four models' rates per million tokens. The Haiku rates are for prompts of 100K tokens or less.
| Model | Input | Output | Cache read |
|---|---|---|---|
| Fable 5.2 | $8 | $40 | $0.20 |
| Opus 5.5 | $4 | $20 | $0.20 |
| Sonnet 5.5 | $2 | $10 | $0.10 |
| Haiku 5.5 | $0.10 | $0.50 | $0.01 |
Of the four models in the table, Opus costs twice what Sonnet costs per token. Output costs five times what input costs on each of them.
Why is the average a weak budget number?#
An average per developer hides the users above $30 on an active day. Anthropic says 90% of users stay below that.
Members who keep reaching their limit may be doing high-value work. Anthropic's Enterprise consumption guide makes this point. When a group keeps reaching its limit, it says to check what the spend is producing before raising the limit. A budget that simply cuts them off saves money and may lose the work.
So never budget Claude Code as one number per head. Budget a range. Write the driver next to each line. Set the date you will check it.
How do you build the budget for 20 developers?#
Build the budget as three lines with the drivers written next to them. Multiplying Anthropic's $150 to $250 per developer by 20 gives $3,000 to $5,000 a month in usage for Enterprise, plus $400 in seats.
The math below uses only Anthropic's published figures. Your pilot will replace them.
| Line | Usage | Seats | Monthly total | Driver to watch (assumed for this example) |
|---|---|---|---|---|
| Low (bottom of Anthropic's average range) | 20 x $150 = $3,000 | $400 | $3,400 | Most work on Sonnet, sessions cleared between tasks |
| Expected (middle of the average range) | 20 x $200 = $4,000 | $400 | $4,400 | Mixed models, some subagents |
| High (top of the average range) | 20 x $250 = $5,000 | $400 | $5,400 | Long sessions never cleared, or Opus as the default model |
The table uses averages, so developers above $30 on an active day do not get their own line. A team with more heavy users than Anthropic's mix can land above the high line.
The seat line is 20 Enterprise seats at $20, which is $400 a month, billed annually as $4,800 a year. The expected line uses $200, the middle of Anthropic's range. Anthropic does not publish a midpoint.
The same team on Team seats looks different. Twenty Standard seats cost $400 a month billed annually. Twenty Premium seats cost $2,000 a month billed annually, or $2,500 billed monthly. A mix of 15 Standard and 5 Premium seats costs $300 plus $500, or $800 a month billed annually. On those seats, usage up to the allowance is already paid for. What you budget on top is usage credits for developers who run past the allowance, billed at API rates. You also budget for the time lost when a developer reaches a limit and waits for the reset. The pricing page describes Premium as 5x the Standard allowance. Watch how often your team reaches the allowance during the pilot.
Then write the line the way a CFO will read it: Claude Code for 20 developers on Enterprise, $3,400 to $5,400 a month. Drivers: the default model, agent use, session habits. Re-forecast at the end of the pilot. That is a line you can explain when it moves.
Which plan fits which team?#
A team where most members stay inside the seat allowance fits Team seats, with Premium seats for the few heavy users. A team where most members' usage varies a lot or runs past the allowance fits usage-based Enterprise or the Console. Seat plans give you a known bill and a ceiling on each person's use, until usage credits are turned on. Usage billing gives you a bill that follows real use, with no ceiling unless you set one. Your pilot decides which fits. On Enterprise and Console routes, compare each developer's pilot spend.
Use this checklist:
- Team, Standard seats: a team of 2 to 150 where most developers use Claude Code for focused tasks, and a fixed monthly bill matters more than usage past the seat allowance.
- Team, mixed seats: a team with a few heavy users. Give Premium seats, with 5x the Standard usage, to the people who keep reaching limits, and Standard seats to everyone else.
- Usage-based Enterprise: a company that needs at least 20 seats and the controls Anthropic lists for Enterprise, such as SCIM, audit logs, custom data retention and role based access. Usage is billed at API rates, so Anthropic's per developer figures are the starting point and your pilot sets the real number.
- Console or a cloud provider: a company that wants every token on one usage bill, or already buys AI through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. On those three clouds, Anthropic's cost page lists three options for per developer tracking: OpenTelemetry (an open source observability framework), a Claude apps gateway, or an LLM gateway.
When should you re-forecast the number?#
Commit to a re-forecast date before you commit to the number. Re-forecast at the end of the pilot, and again one month after the wider rollout. Anthropic tells teams to start with a small pilot group and set a baseline before a wider rollout. Its Enterprise guide says to review group use weekly during the first rollout and monthly after that.
A re-forecast schedule built on that advice:
- The end of the pilot. Replace Anthropic's $150 to $250 range with your pilot's own spread: the typical developer, the heaviest developer, and how many active days each one logged. On Team seats, where usage inside the allowance is not shown in dollars, use the adoption dashboard, and the spend report once usage credits are turned on.
- One month after the wider rollout. New groups bring new habits. Check whether the heavy users are the same people as in the pilot.
- Any change to a driver. Re-forecast when a driver moves: a new default model, agent teams turned on, a new scheduled job, or a new group given Claude Code.
- Every month after that. Compare spend to what it produced. On Enterprise, smart reports break spend down by workstream and type of output, and the Usage page shows how often members reach their limits.
Write the drivers into the budget line itself. When the number moves, the first question will be why. The answer should already be on the page.
What changed for GSC's five engineers?#
Pull requests flagged by GSC's automated code review agent fell from roughly 100% to under 10% in four weeks, and every engineer left with a maintained context file and at least one reusable skill. GSC Enterprises had given five engineers Claude Code, then had Gnar run a four week Immerse cohort, Gnar's mentor-led training for engineering teams, on its own codebases. One of the four things each engineer was assessed on was Diligence, which covers safety, cost and accountability.
GSC is the wholesale distributor behind Grocery Supply Company. Its engineers support a custom ERP on Node.js and AWS, integrations with LANSA on IBM i and DB2, and analytics in Snowflake. Each engineer had adopted Claude Code, and each used it differently. Nobody could say which habits produced good output and which produced rework.
The GSC Enterprises case study records what changed in four weeks:
- Context kept in files. Every engineer left with a maintained context file and at least one reusable skill. Anthropic's cost page also points to a short CLAUDE.md and skills as ways to keep context small.
- A review tool that went quiet. Pull requests flagged by GSC's automated code review agent fell from roughly 100% to under 10%, with no change to the tool. Gnar credits better up front specification and tighter context discipline.
- Inside a normal week. The cohort fit inside a normal workweek, at roughly five to six hours per engineer.
The case study does not report GSC's Claude Code spend, so this post makes no claim about savings. What it shows is the habits. Gnar's AI training for engineering teams is built for a small team that already has the tools but not the practice. If the next step is connecting Claude to your ERP, CRM or databases, start with Gnar Discovery and Design.
