Documentation review, 19 September 2026

AI provider spend controls compared

Spend limits differ by provider, product and account. This reference separates the existence of a control from the ability to read or change it through an API.

This page describes what each provider offers, based on its documentation on 19 September 2026. SpendAssure is read-only: it reports which spending has a limit behind it and does not change these settings. Entitlements and behaviour vary by account. No reviewed source establishes a numerical end-to-end overshoot bound.

Four products, different controls

API spending and enterprise seat products are separate administrative boundaries. A feature on one does not imply the same feature on another.

Documentation-level comparison of four AI products, reviewed 19 September 2026. These are provider capabilities; SpendAssure reads spending data and does not change them.
ProductSpend controlWrite by APIRead by APIControl classification
OpenAI APIOrganisation and project caps, monthly only; 429 on reach; slight overshoot possibleYes, Admin APIYesProvider-capped
ChatGPT EnterpriseWorkspace budget or overage limit; user, group and workspace limitsYes, Spend Controls APIYesProvider-capped
Claude EnterpriseMonthly per-member limits inherited from organisation, seat-tier or group defaults, or overridden per userPer-user overrides only; defaults are configured in claude.ai settingsEffective member limits, inheritance and period spendProvider-capped for covered member usage; no API-writable pooled organisation or group cap established
Anthropic API (Console)Workspace spend limit existsNo, Console onlyNoProvider-capped only as an expiring attestation; otherwise velocity-bounded if all material dimensions qualify, else unbounded

Documentation-level capability inventory. Entitlements, permissions and read or write behaviour vary by account and need checking on each one. No reviewed source establishes a numerical end-to-end overshoot bound; any measurement describes an observation, not a guarantee.

OpenAI API

Documentation reviewed
Usage and cost
Usage and cost endpoints; per-key tracking
Spend cap
Organisation and project caps, monthly only; 429 on reach; slight overshoot possible
Change cap by API
Yes, Admin API
Read cap by API
Yes
Rate controls
Project limits writable (Admin API, Terraform)
Other controls and limits
Model allow and deny lists per project; key management
Control classification

Provider-capped

ChatGPT Enterprise

Documentation reviewed
Usage and cost
Cost API, analytics
Spend cap
Workspace budget or overage limit; user, group and workspace limits
Change cap by API
Yes, Spend Controls API
Read cap by API
Yes
Rate controls
Not applicable
Other controls and limits
Users can request increases natively; approval by API is an open question
Control classification

Provider-capped

Claude Enterprise

Documentation reviewed
Usage and cost
Analytics API, per user
Spend cap
Monthly per-member limits inherited from organisation, seat-tier or group defaults, or overridden per user
Change cap by API
Per-user overrides only; defaults are configured in claude.ai settings
Read cap by API
Effective member limits, inheritance and period spend
Rate controls
Not applicable
Other controls and limits
Approve or deny increase requests by API; requires usage credits enabled
Control classification

Provider-capped for covered member usage; no API-writable pooled organisation or group cap established

Anthropic API (Console)

Documentation reviewed
Usage and cost
Usage in 1-minute, 1-hour and 1-day buckets; cost daily
Spend cap
Workspace spend limit exists
Change cap by API
No, Console only
Read cap by API
No
Rate controls
Read-only by API
Other controls and limits
Disable API key; archive workspace (irreversible, never automated)
Control classification

Provider-capped only as an expiring attestation; otherwise velocity-bounded if all material dimensions qualify, else unbounded

A provider cap is not a guaranteed invoice maximum. Attestations are recorded separately from settings read from a provider API and expire after 30 days by default. Rate limits support a velocity-bounded classification only when every material billable dimension is covered, and a velocity bound is a speed limit, not a ceiling.

Outside this review

These products appear in the source research but were not verified to the same level. Open questions stay open.

Microsoft Copilot Credits

Organisation and user spending policies are recorded in the source research. Whether they can be read or changed through an API is an open question.

Open question

Azure OpenAI, Amazon Bedrock, Google Vertex

Cloud budgets, quotas and model-service controls need separate verification; no provider-wide cap is asserted here.

Not covered by this review

Primary sources

We aim to review provider documentation monthly and when a reported change affects a control. Send corrections to hello@spendassure.com with the product, the source link and the claim concerned. Use these sources to check current availability and the permissions your account needs.

We would be glad to show you a demo

We walk you through the workspace on sample data and discuss what a pilot with your own invoices, exports and contracts would look like.

Agata GrzywaczCo-founder and CEO, e.Lens