Moonshot AI

Kimi K3 is a sparse Mixture-of-Experts large language model developed by Moonshot AI, with 2.8 trillion total parameters and a 1-million-token context window. The model activates 16 out of 896 experts per token using the Stable LatentMoE framework, and is built on two architectural innovations: Kimi Delta Attention (KDA), a hybrid linear attention mechanism that enables up to 6.3x faster decoding in long-context settings, and Attention Residuals (AttnRes), which selectively retrieves representations across model depth and delivers roughly 25% higher training efficiency. Together with refined training and data recipes, these structural advances yield approximately 2.5x better overall scaling efficiency compared to its predecessor Kimi K2. The model applies quantization-aware training from the supervised fine-tuning stage onward, using MXFP4 weights with MXFP8 activations for hardware compatibility. Thinking mode is always enabled at launch, with reasoning effort configurable via the reasoning_effort field.Kimi K3 supports native visual understanding alongside text, accepting image inputs for tasks that combine software engineering and visual reasoning. It targets long-horizon coding, knowledge work, and agentic workflows, and ships in two variants: K3 Max for general chat and agent tasks, and K3 Swarm Max for large-scale parallel processing across many coordinated sub-agents. The model is compatible with the OpenAI SDK via an OpenAI-compatible API. Full model weights are scheduled for release by July 27, 2026 under a Modified MIT license, following the open-weight pattern established by the Kimi K2 model family. A technical report with full architecture, training, and evaluation details is expected to accompany the weights release.

Kimi K2.5 is a frontier-scale multimodal AI model developed by Moonshot AI and released on January 27, 2026. As a significant advancement within the Kimi K2 family, it utilizes a sparse Mixture-of-Experts (MoE) architecture with 1 trillion total parameters (32 billion active per inference) and a massive 256K-token context window. The model features native multimodal integration via a 400M-parameter MoonViT encoder, allowing it to process text, images, and video frames simultaneously. Built for both speed and depth, it offers "Instant" and "Thinking" modes, the latter of which excels at expert-level reasoning, scoring 50.2% on the Humanity’s Last Exam (HLE) benchmark when equipped with tools.The model is released under a Modified MIT License, which remains open-weight but requires attribution for high-revenue commercial entities. It introduces an "Agent Swarm" paradigm capable of coordinating up to 100 specialized sub-agents for parallel workflows, significantly reducing latency in complex research tasks. For vision tasks, Kimi K2.5 demonstrates strong autonomous visual debugging capabilities, where it can inspect its own generated UI outputs against visual specifications to iteratively refine frontend code. This makes it a powerful choice for developers testing automated UI reconstruction, high-fidelity OCR document processing, and multi-step agentic research grounded in complex visual data.