A 1.56 TB download I will never run just changed how I judge open weights. Moo… · @harden_369741 · TwKit

@harden_369741

A 1.56 TB download I will never run just changed how I judge open weights. Moonshot put Kimi K3.1 Pro on Hugging Face: 2.8 trillion parameters total, 104 billion active per token, 1 million token context. I do not own the hardware to run it, and I am guessing you do not either. That used to make me shrug at weight drops. This one is different because the numbers are checkable. Moonshot reports 65.7% on Terminal-Bench 4.0, basically level with Claude Opus 5.5 at 66.4%, and an independent run put it at 67%. Anyone can now verify, distill, or try to beat that score without asking Moonshot for permission. My bet is simple. The next coding agent jump comes from someone remixing a drop like this, not from another closed demo. Who ships that remix first, a big lab or an indie hacker? #OpenWeights #CodingAgents https://www.mlllm.io/articles/3719-moonshot-ai-releases-kimi-k3-1-pro-weights-2-8t-parameter-moe-1m-token-context-and-gated-dsa-sparse-attention/

曝光 23 · 评论 0 · 点赞 0 · 书签 0