Moonshot AI releases Kimi K3, targeting the global frontier with a high-end positioning
AI summary
Moonshot AI released its flagship model Kimi K3 on July 16, with 2.8 trillion parameters. It topped the LMArena Frontend Code Arena with 1679 points, becoming the first open-weight model to surpass Claude Fable 5 and GPT-5.6 Sol on that leaderboard, and ranked 3rd on the Artificial Analysis Intelligence Index (about 57 points). K3 uses a MoE architecture, activating 16 of 896 experts per token, and introduces two in-house technologies, Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), to improve training and inference efficiency, with support for a 1 million token context. Pricing is $3 per million input tokens and $15 per million output tokens, positioning it for the high-end market.
Key Takeaways
- Moonshot AI released its flagship model Kimi K3 on July 16,with 2.
- 8 trillion parameters.
- It topped the LMArena Frontend Code Arena with 1679 points,becoming the first open-weight model to…
- 6 Sol on that leaderboard,and ranked 3rd on the Artificial Analysis Intelligence Index (about 57…
- K3 uses a MoE architecture,activating 16 of 896 experts per token,and introduces two in-house…
Quick Facts
- Published
- Updated
- Source
- Recode China AI(RSS)
- Category
- Model Releases
- AI score
- 80
作者从架构、定价和定位三个层面拆解 K3,指出它是中国首个以能力而非价格竞争的开源旗舰模型。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Moonshot AI releases Kimi K3, targeting the global frontier with a high-end positioning. July 20, 2026. https://new.nat6.net/en/news/o83gusybr5bcou57ox09ei46l/