Kimi Smart Assistant released a public letter early this morning (October 26) acknowledging the poor experience users experienced when using K3.1 Pro recently. The issues mainly lie in two areas: on one hand, the model experiences traffic limits, with response times being blocked during peak hours and task progress being hindered; on the other hand, user quotas quietly disappear without any reason, and they run out before even being used much. The official team admitted that the problems stem from their side, and they should have fixed the bugs earlier.
The recovery plan was released accordingly. Within today, the platform will perform a unified reset for all subscription tiers: as long as the membership is still valid, members will receive a compensation of 25% of their monthly total usage, which is roughly equivalent to a whole week of usage. If someone has already used less than 25%, the progress bar will be reset to zero. The adjustment will be synchronized on the member’s usage page before today ends, and no application is required.
Block the sources of waste
The reduction of the quota is just the first step. Kimi has also started backend configuration. The official statement says that the system has been adjusted, and unnecessary sub-Agent calls are being reduced, along with long output lines. The goal is to enable the same quota to achieve more tasks—directly addressing the previous problem of being “eaten up without any clear explanation”.
Pull the timeline back two days ago; the source of this controversy was the launch of the new model: on October 24, K3.1 Pro, the first model in the K3.1 series, was officially released by Moon’s Dark Side. It is also the most powerful version the company has ever produced—with 3 trillion parameters and the ability to handle 1 million contexts. As soon as the new model went live, users flocked to it in large numbers. Many members complained on social platforms about slow response times and rapid quota deductions, leading to issues with traffic limits and billing. This morning’s apology letter can be considered a splash of calm water on the heat surrounding this topic, and it also reminds AI applications with rapid expansion: no matter how strong the model capabilities are, the stability of the service side and the transparency of billing must not be neglected.