Moonshot AI suspends Kimi K3 subscriptions as demand overwhelms compute capacity
2.8 trillion-parameter model outperforms GPT-5.6, but GPUs can't keep up.
Beijing-based Moonshot AI has paused new subscriptions for its Kimi K3 large language model, citing that demand over the past 48 hours pushed close to the limits of their current computing capacity. The company stated it is adding capacity as fast as possible and will reopen spots in batches, with existing subscribers unaffected.
Kimi K3, a 2.8 trillion-parameter model released last week, has outperformed top US models like GPT-5.6 Sol and Claude Fable 5 on certain benchmarks, including Arena AI's front-end code development ranking. Moonshot plans to release the model's weights by July 27, making it the world's largest open-weight frontier model. However, users report slower operation compared to US alternatives, highlighting the ongoing GPU shortages in China due to US export controls on advanced chips.
- Moonshot AI paused Kimi K3 subscriptions due to demand exceeding GPU capacity within 48 hours of launch.
- Kimi K3 has 2.8 trillion parameters and outperformed GPT-5.6 Sol and Claude Fable 5 on some benchmarks.
- Planned open-weight release on July 27 would make it the largest open frontier model, highlighting China's chip constraints.
Why It Matters
Shows how US chip export controls bottleneck China's AI scaling, even as model quality catches up globally.