Kimi K3 takes the internet by storm; Moonshot AI halts sign-ups after hitting GPU capacity limits
Barely days after unveiling what it calls its most advanced AI model to date, the Chinese startup Moonshot has been forced to temporarily stop accepting new paid subscribers. The reason? Kimi K3 has become too popular. According to the company, demand for its latest AI model has overwhelmed its computing infrastructure.
In a statement posted on X, the company noted: “Over the past 48 hours, demand has pushed our current capacity to the limit. To protect the experience of existing subscribers, we are temporarily pausing new subscriptions and prioritising computing resources for current members. Existing subscribers will not be affected.”
The company adds that current subscribers will experience no interruptions and that new subscriptions will reopen in a phased manner as more computing capacity becomes available.
Changes to the AI subscription model
Alongside the temporary pause, Moonshot has also announced changes to its subscription model. “Going forward, we will split the subscription into two more specialised plans: Kimi Membership, for Kimi Web, App, and Work; and Kimi Code Membership, for programming workflows. This will help us allocate computing resources more precisely and maintain a stable user experience,” the company stated.
The suspension of subscriptions comes just days after Moonshot unveiled Kimi K3 at the World Artificial Intelligence Conference in Shanghai. The company positions this new model as one of the strongest Chinese competitors to leading US AI models, such as those from OpenAI and Anthropic. Additionally, the company plans to release the K3 model weights as open source on July 27, allowing developers to download, customise, and run the model independently.
What is Kimi K3?
Moonshot describes Kimi K3 as its most capable flagship model to date. Based on a 2.8-trillion-parameter “Mixture of Experts” (MoE) architecture, the model features a one-million-token context window and native multimodal capabilities to understand both text and images; furthermore, it is designed for long-reasoning tasks, programming, and AI agent applications.
Unlike traditional chatbots, which are primarily limited to answering questions, Kimi K3 is designed to complete longer, more complex workflows with minimal human oversight. Think of it as an AI assistant capable of executing multi-step tasks, rather than simply responding to a single prompt at a time. Moonshot claims the model can help write code, analyse research papers, process financial data, interpret images, create presentations, and even assist with video editing.
All this is made possible by two new technologies Moonshot calls ‘Kimi Delta Attention’ and ‘Attention Residuals’. The company asserts that these technologies enable the model to process vastly larger amounts of information while using computing resources more efficiently.
A competitor to Claude and ChatGPT
Moonshot positions Kimi K3 as a serious competitor to AI models from OpenAI and Anthropic. The company claims that K3 performs at—or very close to—the level of leading state-of-the-art AI systems across several public benchmarks. It also maintains that Kimi K3 Max outperformed Claude Opus 4.8 (from Anthropic) and GPT-5.5 (from OpenAI) in some of its internal tests, which covered areas such as online research, presentation creation, and financial analysis. However, it should be noted that these results come from Moonshot’s own evaluations and have not yet been independently verified.
This launch takes place amidst intense competition between Chinese and US companies to develop increasingly powerful AI models. Unlike OpenAI and Anthropic, which keep their flagship models closed, Moonshot plans to release Kimi K3 as open source, allowing developers to download and customise it, as well as build applications based on it.