Moonshot AI 已发布其 Kimi K3 模型的全部 2.8 万亿参数权重，使其成为迄今为止最大的开放权重模型。尽管规模庞大且已公开发布，但由于其极高的显存需求（估计超过 1.6TB），在消费级硬件上运行 Kimi K3 实际上是不可能的。此次发布凸显了开放权重运动中的一种紧张关系，即由于部署所需的大量基础设施，模型的商品化可能并不等同于人工智能能力的真正民主化。 AI
影响 凸显了由于大规模基础设施需求，开放权重模型可用性与实际部署可行性之间日益扩大的差距。
排序理由 前沿实验室模型发布，包含系统卡和权重。
AI 生成摘要 · Google Gemini · 来自 7 个来源。 我们如何撰写摘要 →
With Kimi K3 being this good, there is no more ignoring of open-weight models going forward. You are doing yourself a huge disservice if so. Time to start owning every bit of your intelligence stack. Start small if you need to, but start somewhere.
<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://pub.towardsai.net/will-the-us-really-ban-open-weight-models-the-kimi-k3-fallout-b0d0bde06757?source=rss----98111c9905da---4"><img src="https://cdn-images-1.medium.com/max/1400/1*vxgDwwHYV5WynfY-Yk56vg.png…
Moonshot released Kimi K3's full 2.8T weights for free. The move, not the size, is the story: you can't sell 'we have the best model' when the best one is a free download someone else paid to train. Open weights don't democratize AI, they commoditize your competitor's moat. # AI …
<p>On July 27, Moonshot AI delivered on its promise: the full <a href="https://huggingface.co/moonshotai/Kimi-K3" rel="noopener noreferrer">Kimi K3 weights</a> landed on HuggingFace. All 2.8 trillion parameters. 96 shards. 1.56 terabytes. The largest open-weight model ever releas…
<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1v8364f/kimi_k3_weights_now_released/"> <img alt="Kimi K3 weights now released." src="https://preview.redd.it/nlw2mqk9gsfh1.jpeg?width=640&crop=smart&auto=webp&s=45fcb7df195400c006c65e9cde0c40c5436…
<!-- SC_OFF --><div class="md"><p>tldr; we are going to host K3 on A100s (yes, thats correct, we'll try to see if it holds up), H200s & B300s - expect results for A100s & H200s this week while we setup the B300 cluster this weekend & maybe results by next week.</p> <p…
Kimi K3's weights are public — but the license restricts commercial use above a certain scale. "Open" is doing a lot of heavy lifting here. Worth reading the fine print before building anything on it: open weights ≠ open source, and the distinction matters a lot when you're think…