跳到主要內容
AI News HubLIVE
來源內容 · 翻譯待補全1 分鐘閱讀

待翻譯:New Studio M5 Ultra ready for local home AI lab? Compared to OpenRouter prices

文章摘要

AI 服務暫時不可用,以下為來源摘要,待恢復後補全翻譯:Post Log inSign up Post Gabor Herget on X: "New Studio M5 Ultra ready for local home AI lab? I compared it to OpenRouter prices. 512GB will be available from October, and I guess that the price will be around 15-17k. Le…

來源Hacker News AI作者: gaborme
待翻譯:New Studio M5 Ultra ready for local home AI lab? Compared to OpenRouter prices
回報錯誤

更正管道尚未開通,可先複製下方文章資訊留存。

查看更正說明
直接讀正文

AI 服務暫時不可用,以下為來源正文,待恢復後補全翻譯。

Post Log inSign up Post Gabor Herget on X: "New Studio M5 Ultra ready for local home AI lab? I compared it to OpenRouter prices. 512GB will be available from October, and I guess that the price will be around 15-17k. Let's see how much it will cost to run a proper openweight model at home: DeepSeek V4 Pro: 2 Macs," Gabor Herget @gherget New Studio M5 Ultra ready for local home AI lab? I compared it to OpenRouter prices. 512GB will be available from October, and I guess that the price will be around 15-17k. Let's see how much it will cost to run a proper openweight model at home: DeepSeek V4 Pro: 2 Macs, $32–34k Kimi K3 and GLM-5.3: 4 Macs, $64–69k At 100 output tokens/sec, 24/7, compared with OpenRouter prices today, the hardware payback would be: Kimi K3: 16–18 months DeepSeek V4 Pro: 36–39 months GLM-5.3: 56–60 months K3 makes sense for a privacy first fully local setup. Alex Cheema @alexocheema Aug 25 The 512GB M5 Ultra will ship in October with a massive 1.2TB/s memory bandwidth. 50% more memory bandwidth than M3 Ultra, 4.4x more than DGX Spark and M4 Pro. Stacking 4 x M5 Ultra, I expect you'll be able to run Kimi K3 / GLM 5.3 faster than the API (>100 tok/sec). 4:50 PM · Aug 25, 2026177Views

展開要點與分析

文章情報

工程師中級

要點

  • AI 服務暫時不可用,系統已先保留來源內容與降級後設資料。
  • Post Log inSign up Post Gabor Herget on X: "New Studio M5 Ultra ready for local home AI lab? I compared it to OpenRouter prices. 512GB will be available from October, and I guess…

要點與分析由自動化流程生成,可能有誤,請結合原始來源核實。