How Much Does Claude Code Cost a Team? Seat vs API Pricing, Worked Out for 10 and 50 Engineers (2026)

TL;DR

Claude Code costs an engineering team roughly $150–250 per developer per month at API rates. That is Anthropic's own figure: about $13 per developer per active day, $150–250 per developer per month, with 90% of users under $30 per active day across enterprise deployments. On a Claude Team Premium seat the list price is $100 per seat per month billed annually ($125 monthly), but that seat is a usage ceiling, and usage past it is billed at API rates. A 10-engineer team should budget roughly $1,500–2,500 a month in API-equivalent spend; a 50-engineer team, roughly $7,500–12,500. The number that matters more than the total is how much of it is recoverable waste, and no invoice tells you that.

This post is the manager's version of the question: which billing model you are on, what a team-sized budget looks like with the assumptions written down, and where the spend actually goes.

The three ways a team pays for Claude Code

How much Claude Code costs depends first on how your engineers sign in. Anthropic's cost management guide describes three setups, and they meter differently:

Team seats come in Standard and Premium tiers, and the size of a member's Claude Code allowance depends on which one they hold; this post uses Premium, the tier built for daily coding. The three setups can coexist inside one company: per the same guide, "if your organization mixes sign-in methods, each developer is metered according to the one they authenticated with." That alone is why a lot of teams cannot answer the question in the title. Their spend is split across a seat bill, an API invoice, and a cloud invoice, and no single screen adds them up.

What the tokens cost

Under every setup, the underlying unit is the token. Current list prices from Anthropic's pricing page, per million tokens:

Two details on that page matter for budgeting. First, Claude 4.7 and later models use a newer tokenizer that "produces approximately 30% more tokens for the same text," so a budget built on older-model usage under-forecasts. Second, the per-developer average itself moved: in April 2026 Anthropic more than doubled its public estimate from $6 to $13 per active day, and an Anthropic spokesperson attributed it to Claude Code's default moving to a more capable Opus model, not to a price change. Model choice is the biggest lever on the number, and it moves without anyone on your team deciding anything.

A worked budget for 10 and 50 engineers

Assumptions, stated so you can swap in your own:

• Every engineer uses Claude Code as a daily tool.
• API-rate usage per engineer sits inside Anthropic's published $150–250 per month range.
• List prices, no negotiated discount, annual billing where it applies.

Read the Team Premium row carefully. At list price it is cheaper than the API-equivalent average, which is the point of a seat plan, but it is a floor. A seat's allowance is a ceiling on usage, not a promise of unlimited usage, and the engineers who hit it are usually your heaviest and often most productive users. Turning on usage credits so they keep working bills the overflow at API rates. The real monthly cost is $1,000 or $5,000 plus whatever the top of the distribution spends past their allowance.

The distribution is the second thing averages hide. Anthropic's 90th percentile is under $30 per active day, more than double the $13 average. On a 50-person team, a handful of engineers running long sessions, parallel agents, or agent teams, which Anthropic says use roughly 7x more tokens than standard sessions in plan mode, can account for a large share of the total. A per-head budget that ignores that shape will be wrong in a predictable direction.

Why the seat price hides your real spend

On a seat plan, the Claude Code cost doc is explicit: "Usage inside the seat allowance isn't metered in dollars." Finance sees a flat line. That feels like the cost is solved, and it isn't, for three reasons:

1. The overflow is billed at API rates. Usage credits are priced per token, so the moment heavy users go past their allowance, you are on API pricing for exactly the engineers who use the most.
2. The API-equivalent number is the only cross-vendor unit. If part of the team is on Cursor or Codex, a seat price on one tool and a usage bill on another cannot be compared. Pricing every tool's tokens at API rates is the only way to compare what the same work costs on each.
3. A flat number cannot show waste. If 30% of a team's token volume is a context window re-billed on every turn, that is invisible when the bill doesn't move. It still eats the allowance, and the allowance is what makes engineers wait.

This is the same blind spot FinOps closed for cloud spend: one opaque total with no attribution underneath. Applied to AI coding agents, the discipline is called TokenOps.

Where Claude Code spend actually goes

Output is the smallest part. Claude Code "sends your full conversation with every request," per Anthropic's guide, so the cost of a turn is dominated by the context it carries. Here is the arithmetic at list price for one request in a session holding 150,000 tokens of context on Opus 5.5:

• Context served from cache: 150,000 × $0.20 per million = $0.03
• The same context on a cache miss (rewritten at the 5-minute write rate): 150,000 × $5 per million = $0.75, twenty-five times more

On an API key the cache lifetime is five minutes by default, so the first message after a coffee break can cost twenty-five times the one before it. Across a team, the recurring sources are consistent:

• Context carry. Long sessions that are never cleared between unrelated tasks re-bill their whole history on every turn. Anthropic's first recommendation is to clear between tasks. See context efficiency.
• Re-reads. The agent reads the same files again across turns, or scans broadly because the prompt was vague, paying again for content it already had.
• Oversized models. Opus 5.5 costs twice what Sonnet 5.5 does on input and output; Fable 5.1 costs five times as much on output. Anthropic names "Opus left as the default model" as one of the two usual causes of unexpectedly high spend.
• Standing configuration. A long CLAUDE.md and always-on MCP servers load into context on every session whether the task needs them or not.

Together these are what we call recoverable token waste: spend you can remove without removing any of the work.

How to get an honest number for your team

1. Put every sign-in method on one ledger. Seats, API, and cloud usage, priced at API rates so they add up.
2. Attribute it. Per developer, per repo, per workflow. Anthropic's own reporting gives per-user totals (the Team spend report CSV, the Enterprise Analytics API, the Console dashboard), but not which workflow or habit produced the spend.
3. Separate waste from leverage. The goal is not a smaller bill at the cost of less work. It is the same work at a lower cost.
4. Watch the trend, not the invoice. A default-model change or a new agent workflow can double the per-developer number in a month, as Anthropic's own estimate did.

Promptster Teams gives you the real number

Promptster Teams attributes AI coding spend across Claude Code, Codex, and Cursor by repo and workflow (each engineer sees their own figures; managers see team and squad totals), priced at API rates so seat plans and usage bills land on one ledger. It separates recoverable waste (carried context, re-reads, oversized models, redundant runs) from spend that shipped work, and tracks the trend. It reconstructs each session as prompts plus workflow, with code and diffs redacted on the engineer's machine before upload, and it is not developer monitoring.

If the honest answer to "how much does Claude Code cost us?" is "we don't know," that is the problem it solves.

See how Promptster Teams attributes AI coding spend →

Related: What is TokenOps? · Recoverable token waste · Context efficiency · 6 AI spend management tools · TokenOps metrics · Attributing AI coding spend per developer