原帖引用 · AI 翻译
Moonshot AI 在 Hugging Face 发布了拥有 2.8 万亿参数的庞大 Kimi K3.1 Pro 模型权重。
在 Terminal-Bench 上,它与 Claude 5.5 正面交锋
> 该模型以 4 位权重训练,发布时也采用这一精度,避免了 16 位版本所需的 5.6 TB 存储空间。
> Moonshot 将持续增强的漏洞利用能力追溯到其通用训练中期语料库和强化学习环境中隐藏的网络安全内容,并在开放权重前移除了这两部分内容。
> 在 Moonshot 自己的测试中,该模型在 Terminal-Bench 4.0 上得分为 65.7%,接近 Claude Opus 5.5 的 66.4%;但在不使用工具的情况下,它在 HLE-Diamond 上仅得 39.4%,而 GPT-6 Astra 得分为 65.8%。
> 每个 token 仅激活其 1040 亿个参数,93 层中有 69 层采用线性注意力机制,使处理 100 万 token 的上下文成本更低。
> 其修改版 MIT 许可证豁免内部使用,但要求年收入超过 2000 万美元的模型托管公司进行协商,以约束可能以低于该公司自有 API 的价格销售服务的转售商。
原文
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.
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
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