Moonshot AI Releases Kimi K3 Weights and Technical Report

This is genuinely one of the biggest open-weight releases of the year. Moonshot AI just open-sourced Kimi K3. The model weights, full technical report, everything. Here's what you need to know + how to run Kimi K3: The specs → 2.8T total parameters, but only 104B activated per token → Native visual understanding built in, text, images, and video in the same model → 1M-token context window → The world's first open 3T-class model Efficiency inside Kimi K3 Kimi K3 runs on a new architecture (Kimi Delta Attention + Attention Residuals) with a Stable LatentMoE framework that activates just 16 of 896 experts per token. The result: roughly 2.5x the intelligence per unit of compute compared to Kimi K2. How to run it Model weights: Works with vLLM, SGLang, or their own Kimi Code CLI API available [...] Moonshot AI has open-sourced Kimi K3, a 2.8 trillion parameter multimodal model with only 104B parameters activated per token via a Mixture-of-Experts architecture. It features a 1M-token context

Kimi K3 is Kimi’s most capable flagship model to date, with 2.8 trillion parameters. It is built on Kimi Delta Attention (KDA), a hybrid linear attention mechanism, and Attention Residuals, with native visual understanding and a 1M-token context window. It is the world’s first open-source model in the 3-trillion-parameter class, designed for frontier intelligence scenarios including long-horizon coding, knowledge work, and reasoning.For complete benchmarks and case studies, see the technical blog. Kimi is currently working closely with inference partners and open-source maintainers to align technical details and ensure the model launches reliably across the ecosystem. The full model weights will be released by July 27, 2026. More details on architecture, training, and evaluation will be [...] recipes, these structural advances give Kimi K3 roughly 2.5x the overall scaling efficiency of K2, converting compute into capability more effectively.Image 4: Kimi K3 architecture

### 1. Model Introduction Kimi K3 is an open-weight, native multimodal agentic model and Moonshot AI’s most capable model to date. It is a 2.8T-parameter model built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), with native vision capabilities and a 1-million-token context window. It is described by the repository as the world’s first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning. ### Key Features New Architecture: Kimi K3 is built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), and scales up MoE sparsity with a Stable LatentMoE framework that activates 16 out of 896 experts, yielding an approximate 2.5× improvement in overall scaling efficiency over Kimi K2. [...] Geopolitechs # Geopolitechs # Moonshot released Kimi K3 model weights and technical report Geopolitechs's avatar Source: ### Kimi K3 Open Day Thank you for your patience. Today is Kimi K3 Open Day. We are releasi

artificialintelligenceupdater China's AI startup moonshot just released the model weights and technical report of Kimi K3. Kimi K3 is their most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. Also noted that New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, they're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Source: Kimi/X follow @artificialintelligenceupdater for more AI updates. artificial_intelligence_2100's profile picture artificial\_intelligence\_2100 Copied anthrptopic Reply nupourtruly's profile picture nupourtruly

46,352 followers Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. 🌙 Model weights: 🌙 Tech report: 🌙 Tech blog: No alternative text description for this image To view or add a comment, sign in View profile for Michael Tesfaye Hiruy [...] 46,352 followers Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it

Log inSign up ## Post user avatar Kimi.ai @Kimi\_Moonshot Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Kim… Tech report: github.com/MoonshotAI/Kim… Tech blog: kimi.com/blog/kimi-k3 3:14 PM · Jul 27, 202614.2MViews user avatar shirish @shiri\_shh Jul 27 it's gonna fit trust me 120K user avatar Unsloth AI