Claude Console: Workspaces, Keys, Limits 2026

TL;DR

The Claude Console at platform.claude.com is the admin surface for an API organization: workspaces, keys, billing and limits. You get up to 100 workspaces and spend caps from 500 to 200,000 USD. Residency is only us or global. For inference inside the EU, run Claude on Amazon Bedrock.

What is the Claude Console, and what do companies use it for?

The Claude Console is the organization account for the Claude API, not the chat plan on claude.ai. Admins manage members, workspaces, API keys, spend caps and rate limits there. Billing is in USD, prepaid credits or invoice. New models and admin APIs land here first. There is no EU residency control.

Sign in at platform.claude.com. The workspaces guide published on 7 October 2026 keeps the Console separate from Claude Enterprise on claude.ai. Five organization roles gate access, per the Admin API:

Role	Permissions
user	Playground only
claude_code_user	Playground and Claude Code
developer	Playground and API keys
billing	Playground and billing
admin	All of the above, plus user management

Owners and primary owners have every admin permission and can manage admins. The Admin API is unavailable for individual accounts. Create an organization under Settings, Organization first.

We recommend the Console for prototypes, the playground, Claude Code billed through Console, and workloads that do not need EU processing. When personal data must stay in member states, the Console is the wrong production route. Use the GDPR provider comparison.

How do you set up workspaces, roles and keys?

Every organization has a Default Workspace that you cannot rename, archive or delete. Only organization admins create extra workspaces. The default ceiling is 100; archived ones do not count. Scope a key to one workspace, or send anthropic-workspace-id on every request.

The steps follow the workspaces page and the key guide, read on 7 October 2026:

1. Open Settings, Workspaces and click Create workspace.
2. Set a name and colour. Workspace geo is us only today and cannot change after creation.
3. Add members with a workspace role: User, Limited Developer, Developer or Admin. Billing is inherited from the organization billing role.
4. Under Settings, Service accounts, create a service account for CI or production and add it to the workspace.
5. Settings, API keys, Create key: name, expiration, Linked account (you or a service account), optionally one workspace.
6. Copy the sk-ant- secret once into a secrets manager. The Console will not show it again.

Key type	Acts as	Stops working when
Personal key	Your roles	You leave the organization or that workspace
Service account key	The service account	The account is archived or removed from the workspace
Workspace key (legacy)	Nobody; it belongs to the workspace	It expires, is disabled, is deleted, or the workspace is archived
Admin key (sk-ant-admin01-...)	An organization admin	It expires or an admin revokes it

Expiration presets in Authentication: 3 hours, 1 day, 7 days, 30 days, a custom duration, or Never. An organization policy can hide Never. Keys with a lifetime of at least 14 days get an email 7 days before expiry; keys of at least 7 days get one the day before. Disable is reversible. Delete archives the key for good.

The first Claude Code sign-in through Console billing creates a Claude Code workspace automatically. You cannot mint keys there by hand. It is the only workspace with per-user monthly spend limits. Archiving it blocks Console sign-in for Claude Code across the organization. Per-developer cost sits in Claude Code pricing for teams.

A multi-workspace key must send the workspace header on every Messages call. Omit it and the API returns HTTP 400:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "anthropic-workspace-id: wrkspc_01JwQvzr7rXLA5AGx3HKfFUJ" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5-5",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Ping"}]
  }'

Read the workspace ID in Settings, Workspaces, ID column, or from GET /v1/organizations/workspaces?include_default=true. Default Workspace usage and organization-scoped keys often report workspace_id as null. Use the scope field to tell them apart.

Prefer identity-backed keys or Workload Identity Federation over legacy workspace keys. Federation exchanges a token from AWS IAM, Google Cloud, Entra ID, Okta, GitHub Actions or Kubernetes for a short-lived Claude token. Then no sk-ant-api03- string lives in the pipeline. The create-key dialog is in how to create a Claude API key.

What limits and costs apply in the Claude Console?

The Console shows your usage tier and live limits under Settings, Rate limits and Settings, Billing. Start, Build and Scale carry monthly spend caps of 500, 1,000 and 200,000 USD. Custom has no cap. Rate limits are per model class in RPM, ITPM and OTPM.

The rate limits page as of 7 October 2026 publishes the standard tables. New or low-history orgs may start in an Evaluation tier below these numbers.

Tier	Monthly spend cap	Sonnet 5 RPM	Sonnet 5 ITPM	Sonnet 5 OTPM
Start	500 USD	1,000	2,000,000	400,000
Build	1,000 USD	5,000	5,000,000	1,000,000
Scale	200,000 USD	10,000	10,000,000	2,000,000
Custom	no cap	via sales	via sales	via sales

Opus 5 and Haiku 4.5 share the same RPM, ITPM and OTPM numbers as Sonnet 5 on Start, Build and Scale. Fable 5.x sits lower: Start 1,000 / 500,000 / 100,000, Scale 4,000 / 4,000,000 / 800,000. Limits are organization-wide and shared across every inference_geo.

Hit the tier cap and the API pauses until 00:00 UTC on the first day of the next month, unless you raise the tier. The response is HTTP 429 with no retry-after and enforced_spend_limit_reached. A cap you set below the tier returns HTTP 400. Claude Code workspace limits can return 429 with retry-after.

Set workspace limits on that workspace’s Rate limits and Spend limits tabs. You cannot limit the Default Workspace. Workspace values may sit below the organization values, never above. Workspace limits may add up to more than the org; the org cap still wins.

Token prices in the Console are the list prices on the pricing page read on 7 October 2026. Sonnet 5.5 is 2 USD input and 10 USD output per million tokens, Opus 5.5 is 4 / 20, Haiku 4.5 is 1 / 5. Batch halves both sides. US-only inference (inference_geo: "us") on Claude 4.6 and later multiplies every token category by 1.1. Fast mode on Opus 5.5 is 8 / 40 and exists only on the first-party API.

Usage and cost rows in the Console usually appear within 5 minutes, per the Usage and Cost API. Group programmatically by workspace, model or inference_geo. Daily reports allow at most 31 day buckets. Playground traffic has no api_key_id. Claude Platform on AWS does not expose the programmatic usage and cost endpoints; the Console pages still do.

A chatbot that burns 1 million input tokens and 100,000 output tokens on Sonnet 5.5 costs 3.00 USD globally, or 3.30 USD with inference_geo: "us" on a 4.6-or-later model. The same load as a batch job costs 1.50 USD. The full platform table is in Claude API pricing in the EU.

Claude Console vs Bedrock: which route should you pick?

The Claude Console governs the first-party API. Amazon Bedrock governs Claude through IAM in an EU region. The Console gives you a playground, an Admin API, workspace spend caps and Claude Code on Console billing. Bedrock gives you eu. inference profiles and keeps data in EU regions. Foundry has no EU data zone for Claude.

Question	Claude Console	Amazon Bedrock EU
Admin UI	platform.claude.com	AWS console
Auth	sk-ant- or federation	IAM, no Anthropic key
Residency	us or global, workspace geo us only	EU geo IDs, data in EU regions
Spend cap	500 / 1,000 / 200,000 USD by tier	AWS budgets and quotas
Sonnet 5.5 list	2 / 10 USD	10% regional premium from 4.5
Claude Code	dedicated workspace, per-user limits	Bedrock env vars, see setup

The data residency page is explicit: inference_geo applies on the Claude API and Claude Platform on AWS, not on Bedrock and not on Google Cloud. Those platforms pick the region from the endpoint. Foundry replaces inference_geo with Global Standard or US Data Zone (1.1x). Foundry details sit in Azure AI Foundry in Europe.

We recommend Bedrock EU or Google Europe as soon as you process personal data, HR content or customer conversations. Keep the Console for evals, playground, admin automation and data that may leave the EEA. The first EU call is in the Amazon Bedrock EU guide.

How should European teams control residency and cost in the Console?

In the Console you can pin inference to us or global, not to the EU. Set allowed_inference_geos and default_inference_geo on the workspace. Measure the geo that actually ran via usage.inference_geo and the cost report. For EU production, stay on the cloud route.

Workspace residency in the Console, as of 7 October 2026:

curl -X POST "https://api.anthropic.com/v1/organizations/workspaces" \
  -H "x-api-key: $ANTHROPIC_ADMIN_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"name": "evals-global"}'

workspace_geo stays us. An admin can then set data_residency: allowed_inference_geos to ["us"], ["global"] or "unrestricted", plus default_inference_geo. Requests that name a geo outside the allow-list fail. Models older than Claude 4.6 reject inference_geo with HTTP 400. Their usage rows return not_available.

Cost controls that actually work in the Console:

1. Do not run production in the Default Workspace. You cannot put limits on it.
2. One workspace per product and environment, each with a monthly cap and alert thresholds.
3. Set per-model-class rate limits below the org values so one team cannot starve the others.
4. Group the usage report by workspace_id and inference_geo, the cost report by workspace_id.
5. Use Batch for async load (50% off) and prompt cache for repeated prefixes.
6. Cap the Claude Code share under Settings, Workspaces and set per-user limits.

We recommend a hard split: Console workspaces for evals and internally cleared content, Bedrock eu. for anything with personal data. The same split applies to Vertex in Europe, in Claude on Vertex AI in Europe.

FAQ

What does it cost to use the Claude Console?

The Console has no seat price. You pay tokens, cache, batch, optional fast mode, and 1.1x for US-only inference. Sonnet 5.5 is 2 / 10 USD per million tokens, Opus 5.5 is 4 / 20. Start stops at 500 USD in a calendar month, Build at 1,000, Scale at 200,000. Custom agrees limits with sales.

Claude Console vs claude.ai: what is the difference?

The Console is the API organization on platform.claude.com. claude.ai is chat, Team and Enterprise. A paid chat plan does not include API access. Workspaces, Admin API, usage reports and spend caps live in the Console. Claude Enterprise uses different roles and the Analytics API instead of the Admin usage API.

Can the Claude Console keep data in the EU?

No. Workspace geo is us only. inference_geo is us or global only. There is no EU geo. For residency in member states, use Bedrock EU profiles or Google Europe. Foundry has no EU data zone for Claude.

How many workspaces can an organization create?

The default is 100 active workspaces. Archived ones do not count. Ask your account team if you need more. You cannot archive the Default Workspace. Archiving any other workspace archives its keys within seconds and cannot be undone.

What happens when a workspace limit is hit?

A workspace or organization limit you set below the tier cap returns HTTP 400 with invalid_request_error. The tier cap itself returns HTTP 429 with no retry-after until 00:00 UTC on the first of the next month. Claude Code limits can return 429 with retry-after. The anthropic-workspace-id response header names the workspace that was counted.

Should keys expire, or is Never acceptable?

We recommend an expiration of at least 30 days plus your own rotation. Use Never only for keys in a secrets manager that you rotate yourself. An organization policy can block Never. Expired keys return 401 and cannot be reactivated. Create a new key.

Sources

1. Anthropic docs: Workspaces (7 October 2026)
2. Anthropic docs: Get your Claude API key (7 October 2026)
3. Anthropic docs: Authentication (7 October 2026)
4. Anthropic docs: Rate limits (7 October 2026)
5. Anthropic docs: Pricing (7 October 2026)
6. Anthropic docs: Data residency (7 October 2026)
7. Anthropic docs: Admin API (7 October 2026)
8. Anthropic docs: Usage and Cost API (7 October 2026)