https://zoliben.com/en/tags/llama.cpp/Recent content in Llama.cpp on ZoliBen Csupra(Kabra)HugoenThu, 23 Apr 2026 12:00:00 +0000https://zoliben.com/en/posts/2026-04-23-qwen-36-35b-vs-27b-benchmark-results/Thu, 23 Apr 2026 12:00:00 +0000https://zoliben.com/en/posts/2026-04-23-qwen-36-35b-vs-27b-benchmark-results/<p>I finally summed up all the Qwen 3.6 model test results I gathered over the past few days. I compared two models in detail: the <strong>Qwen3.6-35B-A3B</strong> (MoE, hybrid attention/delta) and the <strong>Qwen3.6-27B</strong> (dense, hybrid attention/delta). I ran both with turbo3 KV cache compression on an RTX 4090 as a llama.cpp server.</p> <p>If I had to summarize briefly: the 35B-A3B is <strong>3-4x faster</strong> in everything, but the 27B delivers <strong>better quality</strong>. This is the classic MoE vs. dense tradeoff, just backed by numbers.</p>