RAG not consistently triggered when using GPT-4.1 in Azure AI Foundry

[sugiza-k-eita]

Issue: Inconsistent tool (AI Search / RAG) triggering depending on model used in Azure AI Foundry

I'm experiencing inconsistent behavior when using different models with an agent built in Azure AI Foundry. Specifically:

• When using GPT-4.1, the agent responds normally, but AI Search (RAG) is triggered only about 1 out of 10 times.
• When using GPT-4o or GPT-4.1-mini, AI Search is reliably triggered and RAG works as expected.

I haven’t changed any configuration other than the model, so I’m suspecting this could be a model-specific issue.

### I’d appreciate any insight on the following points:

1. Is GPT-4.1 currently unsupported or unstable with tool usage (e.g., AI Search / RAG) in Azure AI Foundry?
2. Have others observed similar cases where tool calls (like azure_ai_search.query) fail to trigger depending on the model?
3. For production use, would it be more stable to stick with GPT-4o or GPT-4.1-mini for consistent RAG behavior?

[amynic]

Hi @sugiza-k-eita we had a similar question a while back here - could you take a look to see if this helps you from a developer in the community: https://github.com/orgs/azure-ai-foundry/discussions/34