Chinese AI startup Moonshot AI has unveiled Kimi K3, a 2.8 trillion-parameter open-weight model that it says is the world's largest publicly available AI system, delivering performance approaching US frontier labs.
The launch came late on July 16, timed strategically hours before Google's expected Gemini 3.5 Pro launch today at the opening of the 2026 World AI Conference (WAIC) in Shanghai.
Built on a sparse Mixture-of-Experts architecture, Kimi K3 features a 1-million-token context window and ships in two variants: K3 Max for chat and agent tasks, and K3 Swarm Max for large-scale parallel processing. API pricing is set at $3 per million input tokens and $15 per million output tokens.
Moonshot says Kimi K3 incorporates two significant architectural innovations — Kimi Delta Attention and Attention Residuals — which change how information flows across long sequences and model depth. The model also ships with native vision capabilities and always-on reasoning controlled by a tunable reasoning_effort parameter.
Early independent benchmarks from Artificial Analysis place Kimi K3 around the GPT-5.5 and Claude Opus 4.8 tier. Arena.ai ranked it first in web interface building, while Vals AI placed it second overall behind Anthropic's Fable 5.
The model is the first open-weight system to approach the 3 trillion-parameter mark, and Moonshot has promised to release the open weights by July 27 — making it freely downloadable, customizable, and runnable.
Kimi K3's release sent shockwaves through Hong Kong markets, where shares of rival Chinese AI firms Zhipu and Minimax dropped 21.9% and 13.8% respectively.
Backed by Alibaba and Tencent, Moonshot is reportedly seeking $2 billion in fresh funding at a valuation of about $30 billion ahead of a potential Hong Kong IPO.
