Moonshot Resets 25% of Kimi Quotas After Usage Issues

The apology came with a quota reset

Moonshot did not treat this as a demand problem. The apology, posted on the Kimi X account on October 25 Pacific time, October 26 in Beijing, covered two separate failures: subscribers were hitting rate limits that interrupted their work, and the model was consuming quota unnecessarily while doing it. "These were our issues, and we should have fixed them sooner," the company said in the post.

The remedy is a usage-counter adjustment rather than a payment. Every subscriber on every plan gets a one-time reset restoring 25% of their monthly quota, which Moonshot described as equal to one week of usage. On the announced rule, a subscriber who had used 60% of a monthly quota would see recorded usage fall to 35%, and anyone at or below 25% returns to zero. That is our reading of the announced adjustment, not a verified account-level outcome. Moonshot said the usage bars would reflect the adjustment "by end of day PST": the post went up on October 25 Pacific time, so the end of day it named was October 25, not October 26.

What burned the quota

K3.1 Pro is an agent model, and that is where the arithmetic changes. A single request can delegate parts of the job to subagents, and each of those calls draws on the same monthly allowance as the request that started it. Output length draws on it too. When Moonshot says it has adjusted the plan's parameters and is reducing unnecessary subagent calls and output length, it is naming the two dials that decide how much of a fixed quota one task consumes. Neither dial is set by the subscriber.

This is the subscription version of a pattern this site measured at launch. Artificial Analysis found that K3.1 Pro's token prices came down 33% against K3, to $2 per million input tokens and $10 per million output, while its measured cost per completed Intelligence Index task stayed flat at $1.98, because the model spent 55,000 output tokens per task, 14% more than its predecessor. Cheaper tokens, more tokens. That $1.98 figure is an Artificial Analysis measurement of API cost per task, not a record of any subscriber's quota depletion. The billing arrangements are not interchangeable: per-token API pricing makes consumption more directly observable, while a subscription allowance can be depleted by agent activity the subscriber did not explicitly request. That is why the remedy here had to come from the company rather than the bill.

The second capacity apology this year

Reuters notes that Moonshot previously had issues with computing capacity after the K3 launch, and Pandaily reports that in July the company temporarily stopped accepting new consumer subscriptions days after that launch when demand exceeded the compute available. Reuters frames the wider pressure in terms of computing shortages at Chinese labs following US chip export controls. That is broader industry context, not an explanation Moonshot has offered for this incident. This problem is different in kind: not a shortage of capacity, but a model spending its customers' allowance on calls its maker now calls unnecessary. The distinction matters, because the first problem is solved by buying compute and the second is solved by configuring the product, and only one of those was in subscribers' power to anticipate.

What is missing from the announcement is measurement. Moonshot chose a uniform restoration equal to 25% of monthly quota, describing it as one week's usage. It did not disclose how much quota was unnecessarily consumed, how many subscribers were affected or how the adjustment was calculated. Nor has it published the changed rate-limit thresholds, or a before-and-after figure for how much further a quota goes under the adjusted parameters. Subscribers have no way to test the remedy against their usage bars beyond watching whether the next month lasts longer than this one did.

Sources

[1] Moonshot AI (Kimi) on X, October 25, 2026 — the announcement post itself: the apology wording, the one-time reset restoring 25% of monthly quota, the "by end of day PST" timing and the plan-parameter changes; read first-handRead source
[2] Reuters, "Moonshot resets subscriber quotas after Kimi K3.1 Pro launch issues" (Beijing, October 26, 2026; reporting by Liz Lee and Laurie Chen) — the apology and reset terms, and Moonshot's earlier computing-capacity issues after the K3 launchRead source
[3] AI Affairs, "Moonshot AI apologises for Kimi K3.1 Pro rate limits and restores paid quota" (October 26, 2026; reporting Pandaily) — the July 2026 subscription halt precedent and Pandaily's October 24 launch datingRead source
[4] ALFW, "Kimi K3.1 Pro Ranks Fourth on Artificial Analysis. Its Weights Aren't Out Yet." (October 23, 2026) — launch facts, Artificial Analysis pricing and cost-per-task measurements, weights promised by November 4Read source

THE ALFW VERDICT

Moonshot acknowledged the problem and offered subscribers a concrete remedy: a one-time usage reset across every plan. It also says it has adjusted plan parameters to reduce unnecessary subagent calls and output length. Those are appropriate responses, but their effectiveness is not yet established. The 25% restoration is a uniform allowance adjustment, not a published estimate of wasted usage, and Moonshot has disclosed no affected-subscriber count, new operating limits or before-and-after measurements. The larger issue is visibility: on an agent subscription, internal calls and generated output can consume a shared allowance without the customer controlling every step. The next meaningful test is whether subscribers can see, limit and predict that consumption after the changes.