Moonshot AI has temporarily paused new subscriptions for its flagship Kimi K3 model after surging demand pushed the Chinese startup’s GPU infrastructure to capacity, underscoring both the rapid uptake of the system and the operational strain of supporting frontier-scale AI.
Launched on July 16, Kimi K3 is an open-weight large language model with 2.8 trillion parameters and a one-million-token context window, making it the largest open-weight system announced to date and placing it ahead of previous Chinese releases such as DeepSeek’s V4 line. The model is accessible through Moonshot’s consumer app, web platform and developer API, with full model weights scheduled for release on July 27 under an open framework.
Following the launch, Moonshot reported that user demand in the first days of availability had exceeded internal projections, driving GPU utilisation close to the limits of its current infrastructure. In response, the company has suspended new sign-ups for its Kimi subscription service while maintaining access for existing customers, a move aimed at preserving service quality as it works to expand capacity. The pause applies to consumer subscriptions, while the company continues to direct resources to current members and enterprise and developer workloads.
Kimi K3 has drawn attention for its positioning near leading U.S. frontier models on independent benchmarks. Artificial Analysis scores the system at 57 on its Intelligence Index, placing it among the highest-rated models globally and indicating performance in line with top-tier reasoning and coding systems. On task-based evaluations, the model has ranked strongly in areas such as front-end development and long-context work, further driving developer interest.
Moonshot prices Kimi K3’s API at $3 per million input tokens and $15 per million output tokens, reflecting a flagship-tier positioning that is above typical rates among Chinese AI providers. With the open-weight release scheduled for later this month, some future usage is likely to shift to self-hosted and third-party deployments, potentially easing direct load on Moonshot’s infrastructure while extending the model’s reach across the broader AI ecosystem.
Read Article: WordPress Rushes Critical Fix for “wp2shell” Remote Code Execution Flaw
