Why is the agent failing with Llama-3.3 70B Instruct after 25D?

Summary:

Our seeded AI Leave Agent performs flawlessly when configured with GPT‑5 Mini.
However, when we switch only the LLM—keeping the same agent configuration and prompts—the agent fails to behave as expected under Llama‑3.3 70B Instruct.
We can confirm that the tools are invoked, but the agent does not generate any response.

Content (please ensure you mask any confidential information):

Version (include the version you are using, if applicable):

25D

Code Snippet (add any code snippets that support your topic, if applicable):