← 回總覽

Kimi K3 发布:全球最强开源模型,参数达 2.8T

📅 2026-07-17 08:22 meng shao 人工智能 1 分鐘 1106 字 評分: 88
Kimi K3 开源模型 Moonshot AI 大语言模型 AI基准测试
📌 一句话摘要 月之暗面发布开源模型 Kimi K3,拥有 2.8T 参数和 1M 上下文,在多项评测中领先,成为当前最强开源模型。 📝 详细摘要 月之暗面(Moonshot AI)正式发布其最新开源模型 Kimi K3。该模型是全球首个参数规模达到 2.8T 级别的开放模型,支持 1M token 超长上下文和原生多模态,专为长时程智能体编码、知识工作和推理任务设计。官方宣布模型权重将于 2026 年 7 月 27 日前开放。根据推文引用的评测数据,Kimi K3 的整体性能仅次于闭源模型 Claude Fable 5 和 GPT 5.6 Sol,在开源模型中表现最强,稳定领先于 Cla

One Sentence Summary

Moonshot AI releases open-source model Kimi K3, with 2.8T parameters and 1M context, leading multiple benchmarks and becoming the current strongest open-source model.

Summary

Moonshot AI officially released its latest open-source model, Kimi K3. This model is the world's first open model with parameters reaching the 2.8T scale, supporting a 1M token ultra-long context and native multimodality, designed for long-horizon agentic coding, knowledge work, and reasoning tasks. The official announcement states that the model weights will be open-sourced before July 27, 2026. According to benchmark data cited in the tweet, Kimi K3's overall performance is second only to closed-source models Claude Fable 5 and GPT 5.6 Sol, making it the strongest among open-source models, consistently leading over Claude Opus 4.8, GPT 5.5, GLM-5.2, and others. Particularly in the Frontend Code Arena benchmark on Arena, Kimi K3 scored 1679 points, ranking first, surpassing Claude Fable 5 and GPT-5.6 Sol. The tweet includes a link to the official technical blog and a promotional image.

查看原文 → 發佈: 2026-07-17 08:22:11 收錄: 2026-07-17 12:00:50

🤖 問 AI

針對這篇文章提問,AI 會根據文章內容回答。按 Ctrl+Enter 送出。