Anthropic's own figure is the best starting point: across enterprise deployments, Claude Code averages about $13 per developer per active day and $150 to $250 per developer per month, with 90% of users staying below $30 per active day, according to the Claude Code cost documentation (checked October 29, 2026). For a 20-person team using it 18 days a month at the average rate, that is about $4,680 a month in usage. What your team actually pays depends on how you buy it (seats or per-token API billing), which models people default to, and whether anyone can stop a runaway session before it spends a week's budget in an afternoon.
This post covers the pricing options, what drives the bill, how to estimate it before a rollout, and how to set per-developer budgets that stop surprises without slowing anyone down.
Claude Code pricing options
Prices below are from claude.com/pricing as of October 29, 2026. Check the page before budgeting, because plans change.
| Plan | Price | How usage works |
|---|---|---|
| Pro | $20 per month billed monthly, or $17 per month billed annually | Usage included, with plan limits |
| Max | From $100 per month | Higher usage allowance than Pro |
| Team, Standard seat | $25 per seat per month billed monthly, $20 billed annually | Per-seat allowance shared with Claude chat |
| Team, Premium seat | $125 per seat per month billed monthly, $100 billed annually | Larger per-seat allowance |
| Enterprise | $20 per seat per month billed annually, 20-seat minimum | Plus usage billed at API rates |
| Claude Console (API) | Per token | Billed per token to your organization |
| Amazon Bedrock, Google Cloud, Microsoft Foundry | Per token | Billed to your cloud account |
The two families behave differently. On Team and Enterprise plans, each member's Claude Code usage draws from a per-seat allowance that resets on a rolling five-hour window and a weekly window, and admins can turn on usage credits with spend limits for anyone who needs more. On the Console and on cloud providers, every token is billed, so the controls are workspace spend limits, cloud budgets, or a gateway.
What drives the bill
Claude Code charges by token consumption, and token costs scale with context size. Each request carries more than the developer's prompt:
- The whole conversation. Claude Code sends the full conversation with every request, and each tool use sends another request with that batch of tool results. A one-line question in a session that has been open all day still pays for the whole history (at the cached rate, if the cache is warm).
- Model choice. Opus costs more per token than Sonnet. Anthropic's guidance is that Sonnet handles most coding tasks well, with Opus reserved for complex architectural decisions or multi-step reasoning. A switch to Opus also applies to subagents that inherit the session's model.
-
Extended thinking. Thinking tokens are billed as output tokens, and the default budget can be tens of thousands of tokens per request depending on the model. Lowering the effort level with
/effortreduces it. - Cache misses. The first message after a break longer than the cache lifetime reprocesses the full context at full price.
- Parallel work. Agent teams use approximately 7 times more tokens than standard sessions when teammates run in plan mode, because each teammate keeps its own context window. Subagents and scheduled loops add requests on top of the main conversation.
-
Large
CLAUDE.mdfiles and MCP servers. Everything loaded at session start is paid for on every request. Anthropic suggests keepingCLAUDE.mdunder 200 lines and moving specialized instructions into skills, which load only when needed.
The monthly bill is roughly active developers × active days × average daily cost. The variance comes from a small number of heavy sessions, which is why the 90th percentile ($30 per active day) matters more for budgeting than the mean.
Estimate your team's cost before rollout
Use the published averages as a first pass, then replace them with your own numbers after a pilot.
| Input | Example value | Your value |
|---|---|---|
| Developers using Claude Code | 20 | |
| Active days per developer per month | 18 | |
| Average cost per active day | $13 (Anthropic's enterprise average) | |
| Heavy-user cost per active day | $30 (90th percentile) | |
| Share of heavy users | 10% |
With the example values: 18 developers × 18 days × $13 = $4,212, plus 2 heavy developers × 18 days × $30 = $1,080, for about $5,300 a month in usage. On an API or cloud plan that is the bill. On Enterprise it is the usage line on top of seat fees.
Anthropic recommends starting with a small pilot group and using the tracking tools to establish a baseline before a wider rollout. Two weeks of real data from 5 to 15 developers beats any estimate.
Why teams get surprise Claude Code bills
The pattern is usually the same:
-
One shared provider key in a Slack thread, a
.envfile, or a CI secret. Spend cannot be split by person, so nobody knows who spent what. - Monthly budgets only. A runaway loop can spend the whole month in an afternoon.
- No rate limit. A tight agent loop fires requests as fast as the provider allows.
- Opus as the default for everything, including background and simple tasks.
- CI jobs with no cap. A review bot runs on every push with no daily budget and no expiry.
Each of these has a fix that does not slow developers down.
Set per-developer budgets
Where you set limits depends on how your organization accesses Claude Code. The cost docs map it out:
| Your setup | See spend | Cap spend | Per-user reporting |
|---|---|---|---|
| Claude for Teams or Enterprise | Spend report in org analytics | Spend limits in admin settings (organization, group, or member) | Spend report CSV; Enterprise Analytics API |
| Claude Console (API) | Console usage page | Workspace spend limits | Console dashboard, Claude Code Analytics API |
| Bedrock, Google Cloud, Microsoft Foundry | Your cloud billing console | Your cloud's budget controls | OpenTelemetry or an LLM gateway |
The cloud-provider row is the one that most often goes wrong, because cloud budgets alert at the account level and cannot tell one developer from another. An LLM gateway fixes that by giving each developer their own key with its own limits.
With Bifrost, an open-source AI gateway, each developer gets a virtual key with a daily budget (stops runaway loops), a monthly budget, a request rate limit, and a model allowlist. Keys roll up into team budgets, so a request has to pass the developer's daily cap, the team's monthly cap, and any company-wide cap:
curl -s -u "$BF_ADMIN:$BF_PASS" -X POST https://bifrost.internal.example.com/api/governance/virtual-keys \
-H "Content-Type: application/json" \
-d '{
"name": "dev-alice",
"team_id": "<platform-team-id>",
"metadata": {"owner": "alice@example.com", "team": "platform", "cost_center": "cc-1234"},
"provider_configs": [
{"provider": "anthropic", "allowed_models": ["claude-sonnet-4-6", "claude-haiku-4-5"], "key_ids": ["*"], "weight": 1.0}
],
"budgets": [{"max_limit": 20.00, "reset_duration": "1d"}, {"max_limit": 300.00, "reset_duration": "1M"}],
"rate_limit": {"request_max_limit": 120, "request_reset_duration": "1m"},
"is_active": true
}' | jq -r '.virtual_key.value'
The model IDs are examples; use the ones your organization approves. Point Claude Code at the gateway in ~/.claude/settings.json (user scope, never the committed project file):
{
"env": {
"ANTHROPIC_BASE_URL": "https://bifrost.internal.example.com/anthropic",
"ANTHROPIC_AUTH_TOKEN": "sk-bf-your-virtual-key"
}
}
Give every CI pipeline its own key with a small daily budget and an expiry date, separate from any human key:
{
"name": "ci-repo-api-review",
"metadata": {"kind": "ci", "repo": "api", "owner": "platform@example.com"},
"provider_configs": [{"provider": "anthropic", "allowed_models": ["claude-haiku-4-5"], "key_ids": ["*"], "weight": 1.0}],
"budgets": [{"max_limit": 5.00, "reset_duration": "1d"}],
"rate_limit": {"request_max_limit": 30, "request_reset_duration": "1m"},
"expires_at": "2026-12-31T23:59:59Z",
"is_active": true
}
How to pick budget numbers
Do not guess. Run the pilot with generous limits, then set each developer's daily cap at around the 90th percentile of observed daily spend and review after two weeks. Budget rejections and rate-limit hits are a useful signal: a rising count means the limits are too tight or the default model is wrong.
Rate limits for API organizations
If you buy through the Console, Anthropic publishes per-user rate limit recommendations by team size. They apply at the organization level, so individuals can burst above their share when others are idle:
| Team size | TPM per user | RPM per user |
|---|---|---|
| 1 to 5 users | 200k to 300k | 5 to 7 |
| 5 to 20 users | 100k to 150k | 2.5 to 3.5 |
| 20 to 50 users | 50k to 75k | 1.25 to 1.75 |
| 50 to 100 users | 25k to 35k | 0.62 to 0.87 |
| 100 to 500 users | 15k to 20k | 0.37 to 0.47 |
| 500+ users | 10k to 15k | 0.25 to 0.35 |
Cut Claude Code costs without slowing developers down
These habits come from Anthropic's cost guide and cost nothing to adopt:
-
/clearbetween unrelated tasks. Stale context is paid for on every later message. Use/renamefirst so you can/resumethe old session. -
Sonnet by default, Opus on purpose. Switch with
/model. For simple subagent tasks, setmodel: haikuin the subagent configuration. -
Lower the effort level for routine work with
/effort. -
Keep
CLAUDE.mdunder 200 lines and move workflow-specific instructions into skills. -
Disable MCP servers you are not using (run
/mcp), and prefer CLIs likeghwhere they exist. -
Filter verbose output with hooks. A
PreToolUsehook can trim test output to failures only, cutting tens of thousands of tokens to hundreds. - Delegate verbose work to subagents so logs and test output stay out of the main conversation.
-
Check
/usage. On Pro, Max, Team, and Enterprise plans it attributes recent usage to skills, subagents, plugins, and MCP servers, and recommends changes for anything behind 10% or more of your usage.
If your organization pays contracted rates, set the modelPricing managed setting so the dollar figures in /usage, the status line, and OpenTelemetry match your invoice instead of list price.
Install the ai-coding-agent-governance skill
The budget setup, per-agent configuration, and rollout plan in this post are packaged as an open-source agent skill. Your coding agent can run the audit and write the config for you.
- GitHub: github.com/build-with-ai-hq/ai-coding-agent-governance
- skills.sh: skills.sh/build-with-ai-hq/ai-coding-agent-governance
npx skills add build-with-ai-hq/ai-coding-agent-governance
Then ask your agent "we got a surprise Claude Code bill, set per-developer budgets" or "roll out Claude Code to the team with spend limits." When the skill loads, the agent:
- Audits machines and repositories for raw provider keys, agent configs, managed policy, and MCP servers.
- Drafts a one-page policy: approved agents, providers, regions, models, budgets, data rules, MCP servers, and log retention.
- Puts one gateway in front of every agent, with one virtual key per developer and per CI pipeline carrying budgets, rate limits, and model and MCP tool allowlists.
- Writes exact configuration for Claude Code, Codex CLI, Gemini CLI, Cursor, GitHub Copilot, and OpenCode.
- Pins it with Claude Code managed settings delivered by MDM, so developers cannot point the agent elsewhere.
- Sets up CI usage, spend attribution, and adoption metrics, then verifies each control end to end.
For the application side of LLM spend (model routing, caching, and batch APIs), the companion skill is llm-cost-reduction: npx skills add build-with-ai-hq/llm-cost-reduction.
Related reading
- How to Track Claude Code Usage Across Your Team (Per Developer, Per Project)
- Prompt Caching Explained: How It Works on Claude, OpenAI, Bedrock and Gemini (and Why Your Cache Isn't Hitting)
FAQ
Is Claude Code free?
No. Claude Code is included with paid Claude plans (Pro, Max, Team, and Enterprise) or billed per token through the Claude Console or a cloud provider.
Does the Pro plan include Claude Code?
Yes. Pro includes Claude Code with plan usage limits, and Max raises those limits. Usage credits let you keep working past a plan's limit at a spend limit you set.
Is the API or a subscription cheaper for Claude Code?
It depends on usage. At Anthropic's published average of $150 to $250 per developer per month, a seat plan with an included allowance can cost less for steady users, while light or occasional users often cost less on per-token billing. Run a pilot and compare your real numbers.
How do I see what a Claude Code session cost?
Run /usage. The Session block shows token usage by model and a dollar estimate computed from token counts at list price (or at your contracted rates if modelPricing is set). For authoritative billing, use the Claude Console usage page or your cloud bill.
How do I stop one developer from spending the whole team budget?
Give each developer their own credential with a daily cap and a monthly cap. On Team and Enterprise plans, use member-level spend limits for usage credits. On API or cloud billing, use a gateway that enforces per-key budgets.
Top comments (0)