Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]

news.sbsnotebookcheck+1aolChinese AI startup Moonshot AI temporarily halted new consumer subscriptions to its Kimi K3 model in July after surging demand pushed its GPU infrastructure to capacity, marking a milestone moment for Chinese AI as the 2.8 trillion-parameter model became the first from China to top a major coding leaderboard.
Kimi K3, released on July 16, claimed the top spot on Arena.ai's Frontend Code Arena with a score of 1,679 points, surpassing Anthropic's Claude Fable 5 at 1,631 and OpenAI's GPT-5.6 Sol at 1,618. The result represented a 17-place jump from its predecessor, Kimi K2.6, which had ranked 18th on the same leaderboard. The model also ranks third on Artificial Analysis' broader Intelligence Index.aol+2
Kimi K3 is a mixture-of-experts model with a million-token context window and a hybrid linear-attention mechanism called Kimi Delta Attention, designed for efficiency. Its open-weight nature means anyone can freely download and modify it via Hugging Face, a strategy that has become standard among Chinese AI labs seeking rapid adoption.notebookcheck+1
Within 48 hours of launch, demand overwhelmed the company's compute clusters. "Over the past 48 hours, demand has pushed close to the limits of our current capacity," Moonshot AI said in a statement posted on X. "Our GPUs are feeling it." The company announced it had "temporarily paused new subscriptions," with existing subscribers unaffected, and planned to reopen spots in batches as capacity expanded.scmp+1
The Associated Press reported the suspension on July 19, noting the overwhelming demand had pushed capacity close to its limits. The Verge confirmed the pause, adding that Kimi K3 was one of two Chinese models launched that weekend claiming to rival leading U.S. systems.theverge+1
The model's API pricing — $3 per million input tokens and $15 per million output tokens, with cached inputs at $0.30 — undercuts proprietary alternatives from Anthropic and OpenAI Microsoft Corporation . According to data from OpenRouter, token volume from U.S.-based models has fallen from roughly 70% to 30% on the platform, while Chinese model volume has grown to around 60%.aol+1
Gartner projects that global enterprise adoption of Chinese AI models will surge from 5% last year to 50% by 2027, as companies shift toward multi-model strategies that prioritize both performance and cost. Within days of K3's release, Nvidia CEO Jensen Huang posted a letter signed by 25 companies supporting open-weight models — his first post on X.news.sbs+1