{"published":"2026-04-08 09:00:00","html":"<iframe src=\"https://hatenablog-parts.com/embed?url=https%3A%2F%2Fblog-en.fltech.dev%2Fentry%2F2026%2F04%2F07%2Fswebench\" title=\"The Case for Harness Engineering: Achieving SLM SOTA on SWE-bench Verified with a 27B Model (TTS@8 = 74.8%) - fltech - Technology Blog of Fujitsu Research\" class=\"embed-card embed-blogcard\" scrolling=\"no\" frameborder=\"0\" style=\"display: block; width: 100%; height: 190px; max-width: 500px; margin: 10px 0px;\"></iframe>","title":"The Case for Harness Engineering: Achieving SLM SOTA on SWE-bench Verified with a 27B Model (TTS@8 = 74.8%)","description":"Using Qwen3.5-27B without any fine-tuning, we achieved 74.8% (374/500) on SWE-bench Verified \u2014 a benchmark that measures how well a model can fix real OSS issues from GitHub \u2014 by generating 8 candidate patches and selecting the best one. As of April 7, 2026, this is the highest score among local LLMs with fewer than 229B parameters.","version":"1.0","blog_url":"https://blog-en.fltech.dev/","provider_url":"https://hatena.blog","height":"190","width":"100%","categories":["AI","Kozuchi"],"author_name":"kimura-fltech","provider_name":"Hatena Blog","type":"rich","author_url":"https://blog.hatena.ne.jp/kimura-fltech/","image_url":"https://cdn-ak.f.st-hatena.com/images/fotolife/k/kimura-fltech/20260408/20260408121147.png","url":"https://blog-en.fltech.dev/entry/2026/04/07/swebench","blog_title":"fltech - Technology Blog of Fujitsu Research"}