月之暗面在 Hugging Face 发布 2.8T 参数 Kimi K3.1 Pro 权重
Moonshot AI released the massive 2.8T-parameter Kimi K3.1 Pro weights on Huggingface.
月之暗面（Moonshot AI）在 Hugging Face 发布 2.8T 参数的 Kimi K3.1 Pro 模型权重，采用 4-bit 精度训练与发布，避免 16-bit 版本所需的 5.6TB 存储。官方测试中该模型在 Terminal-Bench 4.0 得分 65.7%，接近 Claude Opus 5.5 的 66.4%，但无工具时 HLE-Diamond 仅 39.4%，低于 GPT-6 Astra 的 65.8%。模型每 token 仅激活 104B 参数，93 层中 69 层使用线性注意力以降低 1M token 上下文的处理成本；修改版 MIT 许可证允许内部使用，但年收入超 2000 万美元的模型托管公司需另行谈判。
推荐理由
精选AI 评分 88
原文给出了参数规模、量化精度、基准对比和许可证限制，读者可据此判断开源大模型的开放边界。
主题
查看原帖发布 Kimi K3.1 Pro 的模型权重：https://t.co/tJXcYUeSCj 我们将继续推进开放智能。https://t.co/BbWCJUYYSQ
查看原文
Releasing the model weights of Kimi K3.1 Pro: https://t.co/tJXcYUeSCj We'll keep advancing open intelligence. https://t.co/BbWCJUYYSQ
正文
Moonshot AI released the massive 2.8T-parameter Kimi K3.1 Pro weights on Huggingface.
On Terminal-Bench, its head-to-head with Claude 5.5
> Trained with 4-bit weights, it also ships at that precision, avoiding the 5.6TB a 16-bit copy would need.
> Moonshot traced steadily rising exploit skill to cyber content buried in its general mid-training corpus and RL environments, and removed both before opening the weights.
> In Moonshot’s own tests the model scores 65.7% on Terminal-Bench 4.0, near Claude Opus 5.5’s 66.4%, but without tools reaches only 39.4% on HLE-Diamond, against GPT-6 Astra’s 65.8%.
> Only 104B of its parameters fire per token, and 69 of its 93 layers use linear attention, which makes a 1M-token context cheaper to process.
> Its modified MIT license exempts internal use but makes model-hosting firms earning over $20M a year negotiate, reining in resellers that could undercut the company’s own API.