AI News
Moonshot Apologises and Resets Kimi Quotas After K3.1 Pro Launch Problems
Two days after launching Kimi K3.1 Pro, Moonshot AI apologised on October 26 for rate limiting and unnecessary quota consumption, and offered every subscriber a partial quota reset. The Beijing-based company said on its Kimi X account that it was resetting the monthly quota of all subscribers by 25%, which it said equals one week's usage; subscribers whose usage was already below 25% would be reset to zero. The company's own summary of the cause was blunt: "the problem was ours, and we should have fixed it sooner." For a launch that was supposed to showcase China's most capable open-weight model, the first week has instead become a case study in what happens when demand, configuration and scarce computing meet.
What went wrong: subagents eating quotas
The problems had two parts. Some subscribers hit rate limits while using K3.1 Pro, and others saw their quotas consumed unnecessarily — burned by work they did not ask for. Moonshot said it adjusted parameters to reduce unnecessary subagent calls and output length. Subagent calls are the delegation of tasks among AI agents: when a model breaks a job into pieces and hands them to subordinate agents, each delegation consumes tokens and quota. If the configuration spawns too many subagents, or lets them run too long, a single user request can quietly consume several times the quota it should. Output length works the same way: longer responses cost more, and a model tuned to be thorough will spend a user's budget on verbosity. Both are configuration problems, not model problems — which is why Moonshot could plausibly say the fault was its own, and why the fix was a parameter adjustment rather than a new model. The adjusted usage progress bars were to be displayed before the end of October 25 US Pacific Time, around the afternoon of October 26 Beijing time.
Why this keeps happening to Moonshot
This is not Moonshot's first capacity embarrassment. The company previously had issues with computing capacity after the K3 launch, and the broader context is structural: Chinese AI labs are suffering from computing shortages following US chip export controls. When hardware is scarce, there is less headroom to absorb a launch spike or a misconfiguration — a quota-eating bug that would be an annoyance with spare capacity becomes a service outage without it. The compensation design is also worth noting for what it reveals about the business. The reset covers all active subscription members regardless of tier, at a flat 25% of monthly quota. That is a simple, generous-looking remedy, but it is also the cheapest one that feels fair: it costs Moonshot no cash, it is proportional to what subscribers paid for, and the "reset to zero" rule for users below 25% quietly caps the giveaway for the heaviest users. The apology matters more than the arithmetic. In a market where subscribers can switch to a rival model in minutes, owning a configuration failure quickly — "the problem was ours" — is the fastest way to stop a launch problem becoming a churn problem.
What subscribers should check
If you pay for Kimi, check your usage progress bar now that the adjustment has been applied, and confirm the 25% reset actually landed on your account — the company said the adjusted bars would display by the end of October 25 US Pacific Time, so any account still showing pre-reset numbers after that point is worth a support message. Watch your quota consumption over the next week: the parameter changes to subagent calls and output length should reduce the burn rate, and if your usage still drops faster than your work justifies, that is evidence the fix did not fully land. And keep the launch in perspective. Kimi K3.1 Pro remains the model Moonshot described at launch — a 3-trillion-parameter-class open-weight model with weights due by November 4 — and the quota incident does not change what the model can do. It changes what it costs to find out, which, for a subscription product, is close to the same thing. The weights release date of November 4 has not changed in any of the company's statements about the incident, so subscribers waiting on the open-weight release have no new date to track — only a quota balance to verify.