• 3 min read
Moonshot’s Kimi K3 sparks new China AI buzz
Moonshot AI’s Kimi K3 climbed an AI coding leaderboard and is drawing comparisons to DeepSeek, though it still trails top proprietary US models.
Image: TechXplore
A new model from Beijing-based Moonshot AI is quickly becoming the latest test of how close Chinese labs are getting to the AI frontier. Released on Friday, Kimi K3 has already fueled comparisons to DeepSeek’s breakout 2025 moment, after strong early results and praise from industry observers.
Large language models power chatbots and other tools by processing huge volumes of digital data. Moonshot AI said Kimi K3 is the world’s first open-source model of its size, with roughly 2.8 trillion parameters. In general, more parameters can help a model handle more complex requests, though Anthropic and OpenAI do not disclose parameter counts for their top systems.
Soon after launch, Kimi K3 topped an AI coding leaderboard run by Arena, a platform created by UC Berkeley researchers. Arena also ranked it ninth worldwide for text queries. AFP said its own test found Kimi K3 generated tech-reporting ideas that compared favorably with other chatbots.
That early momentum prompted comparisons to DeepSeek, whose 2025 release challenged assumptions about US dominance in AI. Ethan Mollick, a University of Pennsylvania professor, wrote on X:
“Kimi K3 seems really good, closest to the frontier yet.”
Mollick added that it still “cannot write a good murder mystery,” and said that remains one of AI’s roughest unsolved problems. Tech writer and investor Kevin Xu also pointed to a possible market reaction similar to a “DeepSeek moment.”
Recommended reading
Hassabis says STEM makes you 10x better at AI
Kimi K3 performance against US rivals
Moonshot AI said Kimi K3 showed “frontier-level performance” across its evaluation suite and consistently beat other tested models. But the company also said it still trails the most powerful proprietary models from Anthropic and OpenAI.
Hussein Abbass, a computing professor at UNSW Canberra, said the model appears strong at coding, but its competitiveness across the broader set of foundation-model tasks is still unclear.
“I wouldn’t say they need to be worried. But they shouldn’t stand still.”
Abbass added that performance depends on more than the model itself, including hardware, data centers, and supply chains.
The launch comes as China pushes more aggressively in AI. Kimi K3 follows other closely watched Chinese releases including Zhipu AI’s GLM-5.2. It also arrived during a major tech conference in Shanghai, where President Xi Jinping on Friday called for international cooperation on AI governance.
The broader backdrop remains tense. The United States has restricted exports of advanced chips to China and said earlier this year it was about eight months ahead of China in the field. At the same time, the Trump administration recently delayed the public release of top-end models from Anthropic and OpenAI over concerns they could help hackers break into online systems.
Mollick said the rise of open-weight models like Kimi K3 raises a fresh question: whether US companies will be allowed to speed up how often they release their most advanced systems.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via TechXplore
