Industry Odisha Bureau, July 21: Chinese AI startup Moonshot recently launched the Kimi K3 AI model, and it has broken the internet already. Parent company Moonshot has temporarily stopped accepting new paid subscribers after the model stretched its computing capacity.
Taking it to X (formerly Twitter), Moonshot said the demand over the past 48 hours had pushed its infrastructure close to its limits. In order to ensure a smooth experience for existing customers, it has paused new subscriptions until additional computing resources become available.
The company also said that all existing subscribers will continue to have uninterrupted access. Meanwhile, new memberships will reopen in phases as more GPU capacity is added.
Further, Moonshot also clarified the changes to its subscription plans. Instead of a single membership, it will now offer two separate plans:
- Kimi membership for its web, app, and work services.
- Kimi Code membership for coding related tasks
Moonshot said that the change will help it manage computing resources more efficiently.
Kimi K3 was unveiled at the World Artificial Intelligence Conference in Shanghai. Moonshot describes it as its most advanced AI model so far and plans to release its open source model weights on July 27. This allows developers to download, customise and build applications using the model.
Notably, Kimi K3 uses a 2.8-trillion parameter Mixture-of-Experts architecture. It supports a one-million-token context window and can understand both texts and images. The company designed the model for complex tasks such as coding, research, financial analysis, presentation creation and AI agent workflows.
Moonshot is positioning Kimi K3 as a competitor to AI models from OpenAI and Anthropic. It also claims that the model performs at or near the level of leading AI systems on several benchmarks. However, many of these results might come from the company’s own internal tests and have not yet been independently verified.
Notably, the launch comes right as Chinese and US companies continue to complete in developing more powerful AI models.
