[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-程式碼模型評估":3},{"tag":4,"articles":9},{"id":5,"name":6,"slug":6,"article_count":7,"description_zh":8,"description_en":8},"abd4bf31-c822-44b9-9b69-12fcae056b53","程式碼模型評估",0,null,[10],{"id":11,"slug":12,"title":13,"summary":14,"category":15,"image_url":16,"cover_image":16,"language":17,"created_at":18},"5b168b94-465a-4d72-bbb1-e6577625cb1a","why-coding-benchmarks-are-finally-telling-the-truth-zh","為什麼程式碼基準測試終於開始說實話","BenchLM 的程式碼排行榜顯示，真正有用的訊號只剩 LiveCodeBench 與 SWE-bench Pro；HumanEval 已經不適合拿來選模型。","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778670697069-56o7.png","zh","2026-05-13T11:10:25.586869+00:00"]