Moonshot AI's Kimi K3: A 2.8-Trillion-Parameter Open-Weight Model Rattles Silicon Valley
Chinese startup Moonshot AI has released Kimi K3, an open-weight language model with 2.8 trillion parameters and native vision, built for long-running tasks in coding and knowledge work. Analysts estimate the release has cut the US technological lead in AI to a few weeks. The launch has heightened anxiety in Silicon Valley, as China doubles down on open weights.
AI-processed from SCMP Tech; edited by Hamidun News
Chinese startup Moonshot AI has released Kimi K3, an open-weight language model with 2.8 trillion parameters. According to the South China Morning Post, the release has cut the US lead in frontier AI to "a few weeks" and deepened anxiety in Silicon Valley. The model came out in mid-July 2026.
What model did Moonshot AI release
Kimi K3 is a large language model with 2.8 trillion parameters and open weights, which Beijing-based unicorn Moonshot AI unveiled in mid-July 2026. According to the company, the model has native vision support and is built for long-running tasks: long-horizon programming, knowledge work and analytics.
- Developer: Moonshot AI, a unicorn startup from Beijing
- Size: 2.8 trillion parameters, one of the largest publicly available LLMs
- Open weights: the model can be downloaded and run locally
- Capabilities: native vision, long-horizon coding, knowledge work
- Release date: mid-July 2026
Why it worries Silicon Valley
The release of Kimi K3 has fueled fears in Silicon Valley that US leadership in frontier AI is increasingly fragile. According to the South China Morning Post, the model has narrowed the technology gap between China and the US to "a few weeks", whereas not long ago the lag was measured in months.
The main irritant is not only the model's scale but also how it is distributed. Open weights mean that developers and companies around the world can run Kimi K3 without permission and without paying Western labs for API access, which hits the business model of closed-model players directly.
"Kimi K3 has cut
China's lag behind the US in artificial intelligence to 'a few weeks'," the South China Morning Post writes.
China's bet on open weights
Kimi K3 is part of China's broader push toward open-source AI. By releasing frontier models publicly, Chinese companies are building a global base of users and developers while eroding the commercial advantage of American labs with their closed models.
This is not Moonshot AI's first major model in the Kimi line. But K3, with its 2.8 trillion parameters and native vision, marks the leap that analysts described as shrinking the gap to weeks.
What it means
The gap between Chinese and American frontier AI is closing, and open weights are becoming a geopolitical tool: the more accessible strong models from China become, the harder it is for Western labs to hold on to the premium they charge for closed models.
Frequently asked questions
What is Kimi K3?
Kimi K3 is an open-weight large language model from Chinese startup Moonshot AI. It has 2.8 trillion parameters and native vision support, and it is aimed at programming, knowledge work and analytics.
How many parameters does Kimi K3 have?
According to the company, Kimi K3 has 2.8 trillion parameters, making it one of the largest publicly released language models as of mid-July 2026.
Why does Kimi K3 matter for the AI market?
According to the South China Morning Post, Kimi K3 has cut the US lead in frontier AI to "a few weeks". It also strengthens China's bet on open weights as a way to compete with the closed models of OpenAI and other Western labs.
- Jul 24, 2026Moonshot AI's Kimi K3: A New 'DeepSeek Moment' and the Chips-vs-Architecture Debate
- Jul 24, 2026Moonshot Releases Kimi K3, the Largest Open Model Yet at 2.8 Trillion Parameters
- Jul 24, 2026Moonshot AI's Kimi K3: An Escalation in the Open-Weights Race, and What It Means for AI
- Jul 23, 2026White House Accuses Moonshot of Using Banned Nvidia Chips to Build Kimi K3
Build & Ship Production AI Apps with Vibe Coding
Join Zhemal Khamidun's hands-on Vibe Coding Intensive at hamidun.com/vibecoding — master Claude Code, Cursor, and autonomous agentic workflows.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.