Azure OpenAI & AI Foundry
Azure models run on your own Azure resource, under deployment names you chose, so the credential includes your endpoint and deployments. Pick Azure OpenAI for OpenAI models, or Azure AI Foundry for other models (Llama, Mistral, DeepSeek and so on) deployed in Azure AI Foundry. Both take the same fields.
| Field | What to enter |
|---|---|
| Authentication | API key, or Microsoft Entra ID service principal. |
| API key | One of the resource's keys. Only shown for API key authentication. |
| Tenant ID, Client (application) ID, Client secret | The service principal's Entra tenant ID and application ID (both GUIDs) and its client secret value. Only shown for Entra ID authentication. Forgebench gets access tokens from Entra ID for you. |
| Endpoint | Your resource's endpoint: https://<resource>.openai.azure.com, https://<resource>.cognitiveservices.azure.com, https://<resource>.services.ai.azure.com for Azure AI Foundry, or https://<app>.<env>.<region>.azurecontainerapps.io for an OpenAI-compatible model you host on Azure Container Apps (API key only). Nothing else is accepted. |
| API version | Defaults to 2024-10-21. Change it only if your deployments need another version. |
| Deployments (name = base model, one per line) | Each deployment name and the model it runs, for example my-gpt4o = gpt-4o. Up to 100 per credential. |
Call a deployment by its name with the prefix: my-gpt4o above becomes
azure-my-gpt4o, or azureai-<deployment> for Azure AI Foundry.
Forgebench bills each deployment at its base model's price. If Forgebench has no price for a base model, the save is refused and the error names it. Azure models run only on your credential; there is no shared fallback.
To add the credential, choose Add credential on Admin → API keys → Credentials and pick Azure OpenAI or Azure AI Foundry as the provider. See Model providers.