The standard for open-source big language model AI, Bechmark, has suddenly become === M3 Ultra + DeepSeek R1 === This magical combination. (Not Nvidia and Llama) Whether in Chinese or English, this is currently the most cost-effective and powerful combination in the pan tech community for discussion and testing. For example, Ben Thompson: https://(stratechery.com)/2025/apple-ais-platform-pivot-potential/ Max Weinbach： https://(creativestrategies.com)/mac-studio-m3-ultra-ai-workstation-review/ Even in the M3 Ultra discussion section of HackerNews, DeepSeek is mentioned more than twice as frequently as Llama: https://news. (ycombinator.com)/item? id=43266453