Moonshot AI has paused new sign-ups for its Kimi K3 model after a surge in demand overwhelmed the company's infrastructure within 48 hours of its launch. The model, a 2.8-trillion-parameter system, recently drew attention for outperforming Anthropic's Fable 5 in front-end coding benchmarks while reportedly operating at roughly 40% lower costs than US competitors. However, the unexpected viral demand exhausted the company's available GPUs, forcing Moonshot AI to restrict new access to protect current subscribers.

To manage operational strain, Moonshot AI is restructuring its service access into two distinct tiers: a Kimi Membership for web and mobile applications, and a separate Kimi Code Membership dedicated to coding workflows. This separation is intended to better align hardware capacity with usage patterns, as coding sessions are significantly more compute-intensive than standard interactions. The company intends to reopen new sign-ups in batches as it expands its server capacity.

The capacity bottleneck may be temporary, as Moonshot AI plans to release the full model weights of Kimi K3 on July 27. This strategy allows external organizations to self-host and customize the model on their own hardware, effectively acting as a pressure valve for the company's strained API services. Despite Kimi K3's performance capabilities, the situation underscores persistent industry-wide challenges where compute availability remains a fundamental constraint for scaling frontier models, regardless of geographical location.

Moonshot AI has paused new sign-ups for its Kimi K3 model after a surge in demand overwhelmed the company's infrastructure within 48 hours of its launch. The model, a 2.8-trillion-parameter system, recently drew attention for outperforming Anthropic's Fable 5 in front-end coding benchmarks while reportedly operating at roughly 40% lower costs than US competitors. However, the unexpected viral demand exhausted the company's available GPUs, forcing Moonshot AI to restrict new access to protect current subscribers.

To manage operational strain, Moonshot AI is restructuring its service access into two distinct tiers: a Kimi Membership for web and mobile applications, and a separate Kimi Code Membership dedicated to coding workflows. This separation is intended to better align hardware capacity with usage patterns, as coding sessions are significantly more compute-intensive than standard interactions. The company intends to reopen new sign-ups in batches as it expands its server capacity.

The capacity bottleneck may be temporary, as Moonshot AI plans to release the full model weights of Kimi K3 on July 27. This strategy allows external organizations to self-host and customize the model on their own hardware, effectively acting as a pressure valve for the company's strained API services. Despite Kimi K3's performance capabilities, the situation underscores persistent industry-wide challenges where compute availability remains a fundamental constraint for scaling frontier models, regardless of geographical location.