Moonshot K3: What‘s Our View: China AI Foundation Model | Asia Pacific

研报英文原文证据摘录

Moonshot K3: What‘s Our View: China AI Foundation Model | Asia Pacific

Update

July 17, 2026 11:04 AM GMT

Morgan Stanley Asia Limited+MChina AI Foundation Model | Asia Pacific Gary Yu

Equity Analyst

Moonshot K3: What's Our View Gary.Yu@morganstanley.comLydia Lin +852 2848-6918

Lydia.Lin@morganstanley.com +852 2239-1572

Key Takeaways

K3 has received positive feedback globally, signaling an all-round catch-up of

Chinese LLMs with US leaders in model size, performance, and pricing.

K3, the largest open-weight Chinese LLM so far, may suggest the scaling law still

holds. Its high pricing is also positive to LLM market dynamics. Greater China IT Services and Software

Asia Pacific

We do not view K3 as an overnight miracle but rather as the result of cumulative Industry View In-Line

progress across China’s AI model industry. GLM-5.2, released in June, was already

outstanding.

We expect more Chinese LLMs with larger sizes, higher pricing, and better

performance, with strong global competitiveness.

On July 17, Moonshot released K3, a 2.8T open-weight LLM specialized in long-

horizon tasks, knowledge work, and reasoning, with full weights to be released

before July 27. K3 uses original algorithm innovations—Kimi Delta Attention and

Attention Residual—to improve compute efficiency. In Artificial Analysis ranking, K3

ranks no. 3 globally, only behind two top US LLMs. In Arena.ai's coding leaderboard,

K3 ranks no.1 globally. Its API price is the highest among leading Chinese LLMs,

narrowing the gap with US leaders. With 2.8T parameters, it is also the largest

open-weight Chinese LLM so far.

Exhibit 1: China frontier LLM comparison

AI lab Alibaba MiniMax Z.ai Moonshot DeepSeek Xiaomi

Model Qwen3.7-Max M3 GLM-5.2 K3 V4-Pro MiMoV2.5Pro