A user is experimenting with building a React-style looping agent system using smaller LLMs like Qwen 3.5 9B and Gemma 4, integrated with LangGraph. The agent is designed to handle instructions and images, with tools whose outputs can feed into subsequent tool inputs. The primary challenges encountered include excessive reasoning token generation from Qwen 9B, unstable recursive loops, and truncated or improperly returned outputs after several iterations. AI
RANK_REASON User-generated content on a niche subreddit discussing technical challenges with existing tools and models.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →