Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open model — the first in the 3T-class — featuring native multimodality, a 1-million-token context window, and a Mixture-of-Experts architecture activating 16 of 896 experts [1]. Built on Kimi Delta Attention and Attention Residuals, it achieves approximately 2.5× better scaling efficiency than its predecessor [2]. The model is available via Kimi.com, Kimi Work, Kimi Code, and the Kimi API ($0.30–$15.00/MTok), with full weights to be released by July 27, 2026 [3]. Kimi K3 is designed for frontier intelligence scenarios such as software engineering, knowledge work, and deep reasoning [4].
The model's capabilities include autonomous programming agents, expert-level research assistance, and support for strategic research, pricing and competitive analysis, and stock and crypto investing analysis [5]. With its release, Kimi K3 is poised to make a significant impact in the AI industry, offering a powerful tool for developers, researchers, and businesses alike. The model's open-source nature and competitive pricing ($3 per million input tokens and $15 output) make it an attractive option for those looking to leverage AI in their work [6]. As the AI landscape continues to evolve, Kimi K3 is an important development, showcasing the potential for open-source models to drive innovation and advancement in the field.
Sources
- Kimi K3 Tech Blog: Open Frontier Intelligence (https://daily.dev/posts/lmuaeljpk)
- Kimi.ai (https://x.com/Kimi_Moonshot/status/2077830229968683203)
- Introducing Kimi K3: Open Frontier Intelligence (https://www.reddit.com/r/kimi/comments/1uyc1aw/introducing_kimi_k3_open_frontier_intelligence)
- Kimi API Platform (https://platform.kimi.ai)
- Kimi K2.6 (https://www.kimi.com/ai-models/kimi-k2-6)
- 🚨 China's Moonshot AI has made its Kimi K3 release official ... (https://www.instagram.com/p/Da3UYshAAlR)
