Kimi K3 has gone viral since its release, and it’s strained the resources of its parent company. Moonshot AI announced that it is temporarily pausing new subscriptions to Kimi, saying the model has pulled in far more demand than the company had planned for, and that its GPUs are now running close to their limits.
In a post explaining the decision, the company said the past 48 hours had pushed usage close to the ceiling of its current capacity. To keep things stable for people already paying for the service, Moonshot said it would stop accepting new subscribers for now and put its compute toward serving existing members instead. Anyone who already holds a subscription will continue to have access as normal, the company clarified, and only new sign-ups are being paused.
Moonshot framed the move as protective rather than reactive, saying it did not want the surge in interest to degrade the experience of subscribers who had already committed to the platform. The company said it is working to bring more capacity online and plans to reopen subscriptions in batches as that happens, rather than all at once.
The timing lines up with what has been a breakout week for Kimi K3. The model has been trading blows with the biggest names in the industry on independent leaderboards, placing third behind Fable 5 and GPT-5.6 Sol on the Artificial Analysis Intelligence Index, and topping several coding and agent benchmarks outright. Combined with its status as the largest open-weight model released so far, at 2.8 trillion parameters, the release appears to have drawn in a volume of users well beyond what Moonshot had provisioned for.
Status monitoring services had already flagged signs of the strain before Moonshot’s announcement, with reports of intermittent congestion errors and capacity-related outages on the Kimi API over the preceding days. Investor commentary around the launch had also pointed to a similar dynamic. Atreides Management’s Gavin Baker had argued that Kimi K3 represents a turning point for how compute and value get distributed across the AI industry, a thesis that gets a real-world test the moment a model this capable becomes genuinely popular rather than just well benchmarked.
Alongside the pause, Moonshot used the announcement to lay out a change to how it plans to structure its membership going forward. Rather than a single subscription covering everything, the company will split its offering into two separate plans. Kimi Membership will cover the web app, general app usage, and Kimi Work, while a new Kimi Code Membership will be dedicated specifically to coding workflows. Moonshot said the split is meant to let it match compute allocation more precisely to how people are actually using the product, rather than pooling demand from casual chat use and heavy agentic coding sessions into the same quota.
The company closed its statement by thanking users for their patience while it works through the backlog, without giving a specific timeline for when new subscriptions would reopen or how large the coming batches would be.