Qwen3.8-27B Trades Knowledge for Tool Use
A Reddit user reports that Qwen3.8-27B performs substantially worse than Qwen3.6 on obscure trivia and offline knowledge tests. The model may still be strong for coding and tool-assisted workflows, but users relying on its internal weights for broad factual recall should be cautious.






