Short answer. Stay on Foundry if Azure procurement, Entra ID, regional controls and Microsoft support are contractual requirements. Move to Plugsky if you want an OpenAI-compatible endpoint with a flat monthly bill and cloud, VPC, on-prem or air-gapped deployment. Choose Bedrock or Vertex if you are standardizing on AWS or Google instead. The migration work is mostly auth, deployment names, content filtering and evaluation tooling — not the model call itself.
What Foundry includes
Azure AI Foundry is the umbrella platform: a model catalog, Azure OpenAI models, model router, agent tooling, evaluations and deployment management. Azure OpenAI is the part that hosts OpenAI models inside Azure. The distinction matters because leaving Foundry can mean leaving the platform tooling even if you keep an OpenAI-compatible model call.
| Layer | What it does | Portability risk |
|---|---|---|
| Model endpoint | Chat, embeddings, tools, JSON | Low: the request shape is close to OpenAI |
| Auth | API key or Entra ID | Medium: token acquisition and RBAC differ |
| Deployment names | Your alias for a model deployment | Low: map alias → model ID |
| Content filtering | Azure policy on inputs/outputs | Medium: thresholds and categories differ |
| Evaluations / agents | Foundry tooling and datasets | High: export and re-create |
| Quota / region | Provisioned capacity and residency | High: contractual and capacity-bound |
Azure credit and pricing
Azure's free account typically includes $200 of credit for 30 days plus a set of always-free services. The credit expires, and some AI services are excluded or billed separately. After the credit, model calls are pay-as-you-go, with rates that vary by model, region and deployment type. Check the current Azure free account terms and the Azure OpenAI pricing page before building a budget around the credit.
The alternatives
| Option | Billing | Best for | Main trade-off |
|---|---|---|---|
| Plugsky | Flat monthly | OpenAI-compatible endpoint, 39 models, VPC/on-prem/air-gapped | Not an Azure-native procurement path |
| AWS Bedrock | Per token, AWS billing | AWS-native procurement and IAM | AWS-specific auth and model access |
| Google Vertex AI | Per token, GCP billing | GCP-native procurement and Gemini | GCP-specific auth and quotas |
| OpenRouter | Per token, prepaid | One key across many providers | Provider markup and availability vary |
| Together AI | Per token | Open-weight models and fine-tuning | Per-token bill; check model list |
| Self-hosted | GPU + operations | Full control and residency | Highest operational burden |
Migration: what actually changes
- Auth: replace the Azure key or Entra token flow with the target provider's API key or workload identity.
- Endpoint and api-version: Azure uses a resource endpoint and an
api-versionquery parameter. Most alternatives use a fixed OpenAI-compatible base URL. - Deployment name → model ID: map each Azure deployment alias to a concrete model ID in the new provider.
- Content filtering: re-create policy at the application layer or confirm the new provider's safety controls.
- Evaluations and agents: export datasets and prompts; re-create evaluators and agent tools. This is the largest hidden cost.
- Quota and region: confirm capacity and residency before cutover; do not assume the same limits.
# Azure-style call (illustrative)
POST https://YOUR-RESOURCE.openai.azure.com/openai/deployments/YOUR-DEPLOYMENT/chat/completions?api-version=2024-10-21
# OpenAI-compatible alternative (illustrative)
POST https://api.plugsky.com/v1/chat/completions
Authorization: Bearer YOUR_PLUGSKY_KEY
{"model": "plugsky-pro", "messages": [...]}
When to stay on Foundry
- Azure is a contractual procurement requirement, or you need Microsoft support and billing.
- Entra ID, Azure RBAC and regional controls are already wired into your security model.
- You depend on Foundry evaluations, agent tooling or model router in production.
- Your capacity is provisioned and the migration cost exceeds the savings.
When to move
- You want a predictable monthly bill instead of per-token spend.
- You need one endpoint across several model families without Azure-specific auth.
- You need VPC, on-prem or air-gapped deployment outside Azure.
- You want no monthly token cap on production traffic.
Cost model
Compare the fully loaded cost, not the token rate:
monthly_cost =
model_tokens × rate
+ provisioned_capacity
+ evaluation runs
+ egress / networking
+ engineering time for auth, policy and migration
At low volume, the Azure credit can make Foundry cheaper for the first month. At steady volume, a flat plan can be simpler to forecast. Plugsky's Hobby is $8/month, Starter $20, Builder $60 and Scale $120; Builder and Scale have no monthly token cap, with fair-use concurrency and RPM. See pricing.
FAQ
What is Azure AI Foundry?
Microsoft's platform for building, evaluating and deploying AI applications, including a model catalog, Azure OpenAI, agents, evaluations and deployment controls inside Azure.
Is Azure AI Foundry the same as Azure OpenAI?
No. Azure OpenAI hosts OpenAI models inside Azure; Foundry is the broader platform around it.
How much free credit does Azure give?
Typically $200 for 30 days plus always-free services. The credit expires and some AI services are excluded; check the current terms.
What is the best Azure AI Foundry alternative?
Plugsky for a flat OpenAI-compatible endpoint, Bedrock for AWS-native, Vertex for Google-native, OpenRouter for one key across providers.
Can I migrate without rewriting the app?
Usually the request shape is close, but auth, deployment names, api-version, content filtering and eval tooling need work. Test tool calling and JSON mode before cutover.
Reviewed 20 November 2026. Azure credit terms and Foundry capabilities were checked against Microsoft's public documentation on that date. Cloud terms change; confirm before committing. This page is editorial and does not rank providers by payment.