BREAKING: China's Kimi K3 Just Dropped a 2.8 Trillion Parameter Open AI Model

2026-07-17 · Source: AIM Network · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Expert, quick

Summary

Moonshot AI has released Kimi K3, an open-weight model featuring 2.8 trillion parameters and a 1-million-token context window, making it the world's first open 3T-class model. Utilizing a Mixture-of-Experts architecture with 896 experts (16 active per token), Kimi K3 reportedly surpasses GPT 5.5 and Claude Opus 4.8 on various benchmarks. The model incorporates a Stable Latent MoE framework and a custom Kimi Delta Attention architecture, demonstrating 2.5x better scaling than its predecessor, Kimi K2. It also includes native vision support and is accompanied by new Kimi Work and Kimi Code tools. A full-weight release is scheduled for July 27, 2026, impacting the open-source AI landscape.

Key takeaway

For AI Scientists and Machine Learning Engineers evaluating large language models, Kimi K3's 2.8 trillion parameters and 1-million-token context window offer significant open-source advancement. You should consider its reported performance against proprietary models like GPT 5.5 and Claude Opus 4.8 for demanding benchmarks. Integrate its Stable Latent MoE and Kimi Delta Attention insights into your architecture research. Prepare for the full-weight release on July 27, 2026, to utilize its full capabilities.

Key insights

Kimi K3 introduces a 2.8T-parameter open MoE model with a 1M-token context, setting a new scale benchmark.

Principles

• MoE architectures enable massive parameter counts.
• Custom attention mechanisms enhance scaling.
• Open-weight models drive competitive innovation.

Method

The model employs a Stable Latent MoE framework with 896 experts (16 active per token) and a custom Kimi Delta Attention architecture for improved scaling.

In practice

• Evaluate Kimi K3 for large context tasks.
• Explore Kimi Work and Kimi Code tools.
• Monitor the July 2026 full-weight release.

Topics

• Kimi K3
• Mixture-of-Experts
• Large Language Models
• Open-weight AI
• Context Window
• AI Benchmarks
• Moonshot AI

Best for: AI Engineer, NLP Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Kimi K3 just beat FABLE. — Chinese open-source models, like Kimi K3, are now frontier-class competitors to proprietary LLMs.
• Kimi K2 Thinking: what 200+ tool calls mean for production — Kimi K2 Thinking is an open-source MoE reasoning model capable of 200-300 sequential tool calls.
• The New Frontier: Benchmarking China’s Open-Weight Triad GLM 5.1, KIMI 2.6, MiMo-V2.5-Pro — Open-weight Chinese MoE models are outperforming proprietary Western AI on key developer benchmarks.
• Kimi K3: The Biggest Open Model Ever, Except You Can’t Have It Yet — Moonshot AI's Kimi K3 is a 2.8 trillion parameter model marketed as open, but its weights are not yet publicly available.
• Is the World’s Largest Open-Source AI Model Worth the Hype? — A model's "open-weight" status does not guarantee practical local deployment for massive, trillion-parameter systems.
• What Is Kimi K2.5? Architecture, Benchmarks & AI Infra Guide — Kimi K2.5 is a trillion-parameter open-weight MoE model with multimodal capabilities and agentic modes.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AIM Network.