On this page · 12 sections
- What Anthropic shipped on July 2, 2026
- The spend-limit hierarchy
- Budget alerts: admins at 75% and 90%, users at 75% and 95%
- The Spend Limits Admin API
- Analytics: cost next to the work that caused it
- Model defaults and entitlements
- A rollout order that works
- India-specific considerations
- Limits and caveats
- FAQ
- How eCorpIT can help
- References
Summary. On July 2, 2026, Anthropic shipped a set of cost controls for Claude Enterprise: org-level and per-user spend limits, admin budget alerts at 75% and 90% of the cap, and a Spend Limits Admin API with eight endpoints. Users get their own in-app alerts at 75% and 95% and can request an increase without leaving Claude. Usage and cost data now break down by SCIM group and by individual user, and the Analytics API streams it into Datadog Cloud Cost Management and CloudZero. Spend limits are enforced monthly, amounts are stored in cents (so "50000" means $500.00), and the API is rate-limited to 60 requests per minute per organization. This guide explains how the pieces fit and gives a rollout order that keeps agentic Claude usage from turning into a surprise invoice.
What Anthropic shipped on July 2, 2026
Agentic AI broke the old billing mental model. A chat tool bills roughly per seat; an agent that edits files, runs skills, calls connectors and spins up sub-agents bills per unit of work, and that work is uneven across people and teams. The July 2, 2026 announcement adds richer admin analytics, model-level entitlements, and spend alerts on top of controls Anthropic already offered: spend caps at every level, access and model routing, a usage analytics dashboard with exports, and effort controls.
Two problems get solved. The first is visibility: admins can now see cost by group and by user, with artifacts created, files edited, and skills and connectors used shown next to their cost. The second is control: admins can cap spend per person, alert before anyone gets blocked, and pin cheaper default models to routine work. As Ciro Yamada, Product Director at Nubank, put it in Anthropic's announcement, "Token usage alone doesn't tell you much. What I actually want to see is which skills get run again and again across the org, that's the real signal of value."
The spend-limit hierarchy
A spend limit in Claude Enterprise is not a single global number. Each member has an effective spend limit that resolves from a hierarchy, and the API tells you where it came from through a source field. A group limit is a per-member default, not a pooled group budget: every member inheriting it is gated against their own spend.
| Source level | What it means | Where you set it |
|---|---|---|
| user | A per-user override pinned to one member | Admin API or claude.ai settings |
| rbac_group | The default for a SCIM/RBAC group | claude.ai organization settings |
| seat_tier | The default for a seat tier (for example, enterprise standard) | claude.ai organization settings |
| organization | The organization-wide default | claude.ai organization settings |
| null (unlimited) | No spend limit applies | Default when nothing is set |
| "0" (included only) | Member cannot spend beyond plan-included usage | Explicit cap of zero |
Amounts are strings in the billing currency's minor units. For US dollars that is cents, so "50000" is $500.00 and "75000" is $750.00. The enforced period is currently monthly only, and spend resets at 00:00 UTC on the first of each calendar month. A member's period_to_date_spend can occasionally read "0" when the reading is temporarily unavailable, so treat it as informational rather than a billing source of truth.
Budget alerts: admins at 75% and 90%, users at 75% and 95%
The alert thresholds differ by audience, and getting them right matters because they control who hears about an overage first.
| Audience | Alert thresholds | Channel | What they can do |
|---|---|---|---|
| Admins | 75% and 90% of the org-level spend limit | Admin notification | Raise the cap before anyone is blocked |
| Users | 75% and 95% of their own limit | In-app notification | Request a limit increase from their admin |
When a user hits their ceiling, they click "Request more usage" in claude.ai, which creates a pending increase request. The admin approves it (supplying the new amount, since the request does not carry a requested figure) or denies it. A denied request hides that user's request button for 30 days from the resolution time, though an admin can still raise the limit directly at any point.
The Spend Limits Admin API
For organizations managing limits across many groups, clicking through a console does not scale. The Spend Limits API moves the workflow into scripts. It is available to Claude Enterprise organizations only, not to standard Claude Platform (Console) organizations, and it requires an Admin API key with the read:spend_limits scope for reads and write:spend_limits for writes, passed in the x-api-key header. Usage credits must be turned on for the organization first, which the primary owner does in claude.ai billing settings.
The API exposes eight endpoints across two resources.
| Endpoint | Purpose |
|---|---|
| GET /v1/organizations/spend_limits/effective | List every member's effective limit and period-to-date spend |
| GET /v1/organizations/spend_limits/{id} | Inspect one configured spend-limit row |
| POST /v1/organizations/spend_limits | Set a per-user override (upsert on scope and period) |
| DELETE /v1/organizations/spend_limits/{id} | Remove a per-user override so the member re-inherits a default |
| GET /v1/organizations/spend_limit_increase_requests | List increase requests, filterable by status |
| GET /v1/organizations/spend_limit_increase_requests/{id} | Retrieve one increase request |
| POST /v1/organizations/spend_limit_increase_requests/{id}/approve | Approve a request and write the override in one call |
| POST /v1/organizations/spend_limit_increase_requests/{id}/deny | Deny a pending request |
All eight endpoints share one rate limit of 60 requests per minute per organization, and requests over that return 429. Listing the whole org and setting a $750 override on one user looks like this:
# List every member's effective monthly limit and spend
curl "https://api.anthropic.com/v1/organizations/spend_limits/effective?limit=20" \
--header "x-api-key: $ANTHROPIC_ADMIN_KEY"
# Pin one user to a $750.00 monthly limit (75000 minor units)
curl --request POST "https://api.anthropic.com/v1/organizations/spend_limits" \
--header "content-type: application/json" \
--header "x-api-key: $ANTHROPIC_ADMIN_KEY" \
--data '{"scope": {"type": "user", "user_id": "user_01AbCdEfGh"}, "amount": "75000"}'
One trap worth flagging: setting a limit directly with POST /spend_limits does not resolve a member's pending increase request. Use the approve endpoint if you want to both grant the increase and close the request in one call. Seat-tier, group, and organization-wide defaults cannot be written through this API at all; only per-user overrides are scriptable, and the broader defaults stay in claude.ai settings.
Analytics: cost next to the work that caused it
The analytics dashboard now reports usage and cost by group and by user, filtered by the SCIM groups your IT team already manages, so the breakdown follows your existing org chart. Skills report their own usage and cost, and new endpoints track plugin adoption and artifact creation, which pushes cost attribution past a raw token count toward the work that actually ran.
Claude Code gets two dedicated tabs. Usage shows active developers, session counts and top commands across the org, updated daily. Value estimates productivity lift, cost per commit and annual value, with every formula visible and its inputs adjustable, so finance can argue with the assumptions rather than accept a black box. Analytics chat answers plain-language questions such as "Which teams doubled their Claude usage this month?" and returns charts you can export.
For teams that already run FinOps tooling, the Analytics API is the important part: it pulls Claude usage and cost into Datadog Cloud Cost Management and CloudZero, filterable by date range, team, product or model, so AI spend sits next to the rest of cloud spend instead of in a separate portal.
Model defaults and entitlements
The cheapest control is often not a cap but a default. Model defaults and entitlements let admins set which Claude model new conversations start with across chat, Cowork and Claude Code, so routine work does not silently default to the most expensive model. Admins also control which models are available to specific roles or across the whole organization. Pairing a sensible default model with per-group spend limits does more to flatten a bill than alerts alone, because it changes the unit cost of every routine task rather than reacting after the money is spent. This is the same instinct behind picking Claude Opus 5 effort levels to cut token spend: set the cheaper option as the default path and reserve the expensive one for work that needs it.
A rollout order that works
Turn the controls on in an order that gives you data before you enforce limits.
First, have the primary owner enable usage credits in claude.ai billing settings, because the Spend Limits API will not work without them. Second, connect SCIM so groups and seat tiers map to your real org chart. Third, watch the analytics dashboard for a full billing cycle before you cap anything, so your limits reflect real usage rather than a guess. Fourth, set an organization-wide default limit and a cheaper default model, then layer group and per-user overrides where a team's work genuinely costs more. Fifth, configure the 75% and 90% admin alerts and brief the people who receive them on how fast they must act. Finally, wire the Analytics API into whatever FinOps tool you already run and, if you manage many groups, script the increase-request queue against the Admin API so approvals do not become a bottleneck. Teams that treat this as an AI spend decision framework rather than a one-off setting keep the bill predictable as usage grows.
India-specific considerations
For Indian enterprises and global capability centres running Claude, three points matter. Billing is in US dollars, so a $500 monthly org cap is roughly Rs 42,000 to Rs 44,000 at mid-2026 exchange rates; budget in rupees but read the API amounts as USD cents. Second, the Analytics API exports usage and cost data to external tools such as Datadog and CloudZero, so that flow is a data-processing path: usage metadata about your people leaves Claude for another system, and under the Digital Personal Data Protection Act, 2023 (DPDP) you should map who that data identifies and where it lands before you turn the export on. Third, per-user spend visibility surfaces individual activity, which is useful for cost control but is also employee data, so handle it under the same access and retention rules you apply to other workforce analytics. The wider discipline is the same one behind any cloud FinOps programme for Indian teams: attribute spend, alert early, and keep the data flows compliant.
Limits and caveats
The controls are new, and a few edges are worth knowing. The enforced period is monthly only for now, so you cannot set a weekly or quarterly cap through the API yet. A group limit gates each member individually and is not a shared pool, so a five-person group on a $200 limit can spend up to $1,000 combined, not $200. The period_to_date_spend field is informational and can momentarily read zero. And the Admin API writes only per-user overrides, so your organization, seat-tier and group defaults still live in claude.ai settings. None of this is a blocker, but each one changes how you script and how you explain the caps to teams. It is the same lesson as any enterprise AI credit cost-control rollout: the defaults and the exceptions both need an owner.
FAQ
How eCorpIT can help
eCorpIT helps enterprises and global capability centres put AI spend under real governance, not just a monthly invoice. We map your SCIM groups to sensible default models and spend limits, configure the 75% and 90% admin alerts, and script the Spend Limits Admin API so approvals and overrides scale across teams. We wire the Analytics API into the Datadog or CloudZero setup you already run and fold Claude spend into a cloud FinOps managed service so it stays predictable. As a CMMI Level 5 and ISO 27001:2022 certified organisation, we design the usage-data flows aligned with DPDP Act 2023 requirements. Tell us your setup at /contact-us/.
References
_Last updated: July 29, 2026._
