Azure API Management vs Portkey as an LLM gateway

Choose Azure API Management if your company is standardized on Azure, already runs APIM, and wants token limits, load balancing and logging in front of model deployments it owns, including Claude through the Anthropic Messages API on the v2 tiers. Choose Portkey if you want a dedicated AI gateway with guardrails, observability and audit logs on an Enterprise contract, or a data plane inside your own VPC.

With both, you still deploy or contract every model yourself. Microsoft describes the AI gateway as a set of capabilities of API Management's existing gateway, and Portkey's site now presents it as Prisma AIRS AI Gateway from Palo Alto Networks.

Last updated 9 November 2026. Facts about Azure API Management and Portkey come only from their own public pages, listed under Sources.

Azure API Management and Portkey side by side

Azure API Management compared with Portkey
Azure API ManagementPortkey
What it isMicrosoft's API gateway with AI gateway policies; not a separate offeringAn AI gateway with observability and guardrails; open-source core (MIT), SaaS, or a hybrid data plane in your VPC
PricePer gateway unit by tier (classic and v2), or per API operation on ConsumptionFree developer tier; Production $49 a month for 100k recorded logs, plus $9 per additional 100k; Enterprise custom
Model contractsModels you deploy: Microsoft Foundry, Amazon Bedrock, Google Vertex AI or self-hostedYour own provider keys, stored as Integrations in the Model Catalog
API schemasOpenAI Chat Completions and Responses; Anthropic Messages on v2 tiers; Google Vertex AIOpenAI-compatible API across the providers you connect
RoutingRound-robin, weighted, priority and session-aware load balancing, with a circuit breakerConditional routing, fallbacks and load balancing you configure
Token limitsTokens per minute, or a token quota per hour, day, week, month or year, on any counter keyUsage and rate limit policies on the Enterprise Self Hosting plan
Processing in EuropeRuns in the Azure regions you deploy to, in front of backends you deployEU-hosted SaaS region: not stated; hybrid data plane in your VPC on Enterprise
Prompt loggingOptional per API: prompts and completions to Azure Monitor, in 32 KB chunksRequest logs are recorded; Production is priced per recorded log
Zero data retentionDepends on each model deployment behind itDepends on each provider contract you bring
Single sign-onMicrosoft Entra ID for the developer portalSSO on Enterprise; SCIM provisioning (Okta, Entra ID) in its enterprise docs
Usage and auditToken metrics per consumer in Application Insights and a built-in analytics dashboardAudit logs on Enterprise
From each vendor's own pages, checked 9 November 2026. "Not stated" means we found no public statement.

Where they differ most

What you run

API Management is an extension of the API gateway you may already operate, configured with policies such as llm-token-limit and llm-emit-token-metric. Portkey is a dedicated AI gateway with a SaaS plan, an open-source core and a hybrid deployment where the data plane runs in your VPC.

Claude next to GPT

API Management governs language model APIs that follow the OpenAI, Anthropic Messages or Vertex AI schemas, with Anthropic Messages currently supported in the v2 tiers, and models deployed in Microsoft Foundry or providers such as Amazon Bedrock. Portkey connects the providers whose keys you add to its Model Catalog.

Limits and logs

API Management limits tokens per consumer and can log prompts and completions to Azure Monitor when you enable it per API. Portkey records request logs, prices its Production plan by recorded log, and keeps audit logs and usage policies for Enterprise.

Choose Azure API Management or Portkey

Choose Azure API Management if

  • Your platform team already runs API Management and wants AI traffic under the same policies.
  • Your models are deployments you own in Microsoft Foundry, Amazon Bedrock or Vertex AI.
  • Token quotas per app are enough, and you do not need budgets in currency.

Choose Portkey if

  • You want a dedicated AI gateway with guardrails and observability, separate from your API platform.
  • Policy requires the data plane to run inside your own VPC.
  • You are ready for an Enterprise contract to get SSO, audit logs and usage policies.

A third option for companies

Both gateways sit in front of model deployments and contracts you already hold. If your company is on Azure but also uses Claude, and would rather buy the models and the controls in one place, Standard Compute is a managed gateway that supplies Claude, GPT, Gemini and open models itself, and your people sign in with their Microsoft work accounts.

Cost per step

Each step goes to the cheapest model that can do it and the cheapest provider fast enough to serve it.

Models

Claude, GPT and Gemini alongside open models such as GLM, DeepSeek, MiniMax and Kimi, through one API.

Processing in Europe

A Europe rule per company, team or key: every request, including retries and fallbacks, is served only by approved European inference endpoints, or refused before your prompt is sent.

Zero data retention

Required per company, team or key, and Standard Compute never trains on your prompts or outputs.

Single sign-on

Microsoft Entra ID or any SAML provider, and Claude Desktop and Claude Code sign-in with a Microsoft work account.

Budgets

A fixed monthly budget per person and app that never overruns, and spending caps per key and for the whole company.

Allow-lists

Approved providers per company, team or key, and approved models per key. Fallbacks never route around them.

Audit log

Admin changes with before and after values, and a 12-month export of every request's model, provider and region rule.

Sign our Data Processing Agreement in the dashboard, and see every sub-processor with its purpose and location.

Create a company workspaceor write to contact@standardcompute.com

Questions

Can Azure API Management be used as an LLM gateway?

Yes. Microsoft documents AI gateway capabilities in all API Management tiers: importing language model APIs, token limit policies, semantic caching, load balancing with a circuit breaker, content safety checks and logging of prompts and completions to Azure Monitor.

Does Azure API Management support Claude?

Microsoft lists the Anthropic Messages API among supported schemas, currently in the v2 tiers, for models deployed in Microsoft Foundry or providers such as Amazon Bedrock. You still deploy and pay for the Claude models yourself.

Does Azure API Management set spending budgets?

Its llm-token-limit policy sets tokens per minute or a token quota per hour, day, week, month or year on any counter key. Budgets in currency are not stated on Microsoft's AI gateway page.

Does Portkey offer EU data residency?

Portkey does not state an EU-hosted SaaS region. It offers hybrid deployment, where the data plane runs in your VPC, and says its Production plan is not recommended for organizations requiring data residency guarantees.

Is there a gateway for Azure companies that includes Claude and the controls?

Standard Compute is one API for Claude, GPT, Gemini and open models. Your people sign in with Microsoft Entra ID, and you get a Europe rule, zero data retention, budgets per key and for the company, allow-lists and an audit log, with each step routed to the cheapest model that can do it.

Sources (checked 9 November 2026)

  1. Microsoft Learn: AI gateway capabilities in Azure API Managementlearn.microsoft.com
  2. Microsoft Learn: logging for language model APIslearn.microsoft.com
  3. Microsoft Learn: API Management region availabilitylearn.microsoft.com
  4. Microsoft Learn: developer portal sign-in with Microsoft Entra IDlearn.microsoft.com
  5. Azure API Management pricingazure.microsoft.com
  6. Portkey pricingportkey.ai
  7. Portkey plan feature comparisonportkey.ai
  8. Portkey Model Catalogportkey.ai
  9. Portkey audit logsportkey.ai
  10. Portkey budget policiesportkey.ai
  11. Portkey hybrid deploymentportkey.ai
  12. Portkey gateway on GitHubgithub.com
  13. Portkey SCIMportkey.ai