Hello Nermes,

Thanks for sharing your feedback. We're working on improving this in the next release by reducing the agentic thinking steps, which should fix these slowdowns. For now, we recommend using non-thinking models like GPT-4.1 to avoid these issues.