Newer Qwen models are worse at summarization?

In three linesUser reports Qwen 3 (30B) outperforms newer models on summarization tasks evaluated by LLM judge, followed by Gemma 4. Suggests recent Qwen versions may be optimized for agentic tasks rather than synthesis.

Summary generated by Claude — human-verified