The issue maybe from differences in prompt clarity or agent configuration, as GPT-4o-mini can be inconsistent with function calling unlike GPT-4o, so ensure prompts are explicit and compare settings with working agents..