Moonshot AI upgrades K3 with stronger coding focus
SingTao · 1 SOURCESabout 2 hours ago2 MIN
Summary
Moonshot AI has unveiled Kimi K3.1 Pro as the first major upgrade to its K3 flagship model, positioning it for complex professional tasks including programming, 3D content creation and financial research. The company said the model has 2.8 trillion total parameters, uses a mixture-of-experts architecture with 104 billion activated parameters per inference, and supports a one million-token context window. Moonshot AI also released a faster Kimi K3.1 Pro Highspeed version, which it said delivers similar intelligence while generating output about two to three times faster. The company plans to open the model’s full weights by November 4, while both versions are already available through Kimi’s website, app, Kimi Work, Kimi Code and API.
Key Points
- Moonshot AI said Kimi K3.1 Pro is aimed at demanding professional use cases, especially coding, 3D creation and financial research workflows.
- The model has 2.8 trillion total parameters, activates 104 billion per run under MoE, and supports a one million-token context window.
- In Terminal-Bench 4.0, Kimi K3.1 Pro scored 65.7%, above OpenAI’s GPT-6 Astra at 58.2% and close to Claude Opus 5.5 at 66.4%.
- In Moonshot AI’s internal Kimi Code Bench 3.0 Long-horizon test, the new model scored 61.2%, up from 55.95% for K3.
- Moonshot AI recently completed its final private fundraising at about a US$50 billion valuation and is expected to seek a Hong Kong listing in the first quarter next year.
Why It Matters
For Hong Kong readers, the expected first-quarter listing could make Moonshot AI one of the most closely watched technology flotation candidates in the city’s market. The upgrade also shows how Chinese AI developers are competing not only with US model makers but also with domestic rivals such as DeepSeek and Zhipu in coding and finance-related applications that matter to enterprise users.