Community
Qwen3.8 27B with thinking matches Claude Opus 5.5 on a Korean RAG benchmark in one local test
A single temperature-0 run on Allganize's Korean RAG set scored Qwen3.8 27B at 77.3% and Opus 5.5 at 76.6%, a statistical tie. Thinking off drops Qwen about 6 points, and the author notes the results are not yet repeated across runs.
Every outlet that ran it 1 article
r/LocalLLaMACommunityLead
qwen3.8 27B on my 3090 basically tied Opus 5.5 on a Korean RAG benchmark, which I did not expect