How to budget Claude Code for a team

Claude Code spend is hard to predict before a team starts, and it varies a lot from one developer to the next. A budget that holds up has three parts: a starting estimate, a cap that stops a bad month turning into a bad quarter, and a review against what people actually used. Every Anthropic figure below was read on 2026-10-02.

1. Start from Anthropic's numbersCopy Anchor Link

Anthropic's cost guide gives a published baseline across enterprise deployments:

• About $13 per developer per active day on average.
• $150–250 per developer per month.
• Below $30 per active day for 90% of users.

A first estimate is developers × active days × cost per day. For a team of 10 developers each active on 20 working days a month (an assumption; use your own calendar):

Anthropic's $150–250 a month implies fewer active days, about 12 to 19, which puts the same team at $1,500–2,500. Budget the average and cap near the high end. A small group of heavy users, long Opus sessions or automation can sit above the 90th percentile, so a single average hides most of the risk. Anthropic's own advice is to run a small pilot first and measure it before a wider rollout.

2. Pick how the team paysCopy Anchor Link

The billing model decides which controls you get (Manage costs):

With seats, the allowance is the default ceiling, so spend beyond the seat happens only where you turn on usage credits. Per-token billing tracks usage exactly, so a quiet month costs less and a busy one more. Prices for seats are on claude.com/pricing and per-token rates on Anthropic's pricing page. If a developer signs in another way, such as a personal Max plan, their usage is metered there and never reaches your organization's reports.

3. Set the capsCopy Anchor Link

Teams and EnterpriseCopy Anchor Link

The seat allowance is the default ceiling. Members go past it only when you turn on usage credits, and then only up to the limits you set (consumption guide):

• Per member, overriding any group limit, higher or lower.
• Per group: every member inherits the same individual limit. Anthropic recommends starting here, with groups by job function rather than org chart.
• Pooled group budgets (Enterprise, beta): one shared monthly amount a group draws from, for a team with a fixed budget and uneven usage. When the pool runs out, the whole group pauses.
• Organization-wide, as a hard ceiling. Hitting it stops everyone at once, so keep it above the sum you expect.

On Enterprise, admins are alerted at 75% and 90% of an organization limit; pools alert at 50%, 75%, 95% and 100%. With no limit anywhere, a member's spend is not capped.

Claude ConsoleCopy Anchor Link

Each usage tier carries a monthly spend cap ($500 on Start, $1,000 on Build, $200,000 on Scale, none on Custom), and you can set your own lower limit on the Billing page (rate limits). Claude Code gets its own "Claude Code" workspace the first time someone signs in with a Console account. Give that workspace a spend limit and, if your organization has custom rate limits, a workspace rate limit, so Claude Code cannot crowd out production traffic.

For rate limits, Anthropic suggests per-user tokens and requests per minute by team size, falling as the team grows because fewer people use it at the same moment (Manage costs):

Cloud providers and gatewaysCopy Anchor Link

On Bedrock, Google Cloud or Foundry, Anthropic's dashboards do not see the usage. A self-hosted Claude apps gateway can cap each developer by day, week or month, and blocks them until the period resets on a UTC boundary or an admin raises the cap. An LLM gateway that tracks spend per key is the other option.

Scripts and CICopy Anchor Link

A headless run can cap itself: claude -p --max-budget-usd 5.00 "…" stops once the run's estimated spend, subagents included, reaches the amount. Use it on anything scheduled, where nobody is watching the meter.

4. Lower the cost per taskCopy Anchor Link

Caps limit the damage; habits lower the bill. The levers with the most effect, from Anthropic's cost guide:

• Set the default model. Opus left as everyone's default is a common cause of unexpectedly high spend. Match the model to the job, and use Sonnet for agent teammates.
• Lower the effort level for routine tasks. Thinking tokens are billed as output.
• /clear between unrelated tasks. Every request re-sends the whole conversation; see cache tokens explained.
• Keep agent teams small and shut teammates down when their work is done. Each one runs its own context window.
• Report at contracted rates. If you pay less than list price, the modelPricing managed setting makes /usage, the status line and OpenTelemetry show your rates.

5. Review against real usageCopy Anchor Link

Review weekly while the team ramps up, then monthly. Look at who is near their limit before raising it: Anthropic notes that members at their limit are often your strongest adopters, and a lower effort level or clearer model guidance can matter more than a bigger cap.

The review needs one view of everyone's usage, and that is where mixed accounts break down: a contractor on their own plan, a developer on a personal Max seat and a cloud session that leaves no files behind all fall outside your organization's reports (SessClone counts cloud sessions on Pro and Max only, because they need Claude's API credentials). SessClone's Collector runs on each machine and reports every Turn to your Org whatever account it was billed to, so Costs shows one estimate per Member, Project, Device and day. An Enterprise Org can set its own per-model rates so the estimates match a negotiated contract. It tracks spend; the caps stay with Anthropic, your cloud or your gateway.

Track costs per developer and team covers the setup, and SessClone vs ccusage vs Anthropic's analytics compares the options. Hosted plans are on the pricing page; self-hosting is free at any size. Create an Org to start.