Kimi Offers a 25% Monthly Quota Reset After K3.1 Pro Problems

Kimi announced a one-time restoration of 25% of monthly quota for every subscriber on every plan after acknowledging rate limits and unnecessary usage consumption on K3.1 Pro. The October 25 announcement also promised fewer unnecessary calls to subsidiary agents and shorter output, changes intended to make an allowance last longer. Kimi

The credit applies to consumed monthly allowance. Under the announced rule, usage below 25% returns to zero. Applied as a deduction of 25 percentage points, an account showing 80% used would show 55%; one showing 10% would show zero. Those are illustrations of the stated adjustment, rather than observations from subscriber accounts. Kimi

K3.1 Pro had reached Kimi Code on October 23, according to the product’s release notes. It is available on Plus and the older Moderato tier or above, with larger context and the Highspeed variant reserved for higher tiers. The release expanded access to a model that can work through coding, research and other multistep jobs, while leaving several separate usage rules in place. Release

Monthly credit and short-term limits are separate

Kimi’s current membership documentation separates total monthly quota from a rolling five-hour rate window. New plans have no weekly quota window; older Andante, Moderato, Allegretto and Allegro subscriptions retain their existing weekly rules. Both groups still have a monthly ceiling. Reaching that ceiling freezes Kimi Code usage until the monthly reset or an upgrade, according to the documentation. Plans

That distinction explains why a credited balance does not describe every interruption. Kimi’s error guide separately identifies five-hour exhaustion, weekly exhaustion on legacy plans, monthly exhaustion, concurrent-request limits and temporary engine overload. Several return the same HTTP 429 status. The accompanying message identifies whether a customer has spent an allowance, sent too many requests together or encountered a busy service. Errors

For account-quota exhaustion, the guide says repeating the request is pointless; the relevant window must reset or the account needs additional access. Permission failures are different again. A request for a model or context length outside a plan can return HTTP 403. Waiting for a five-hour window to recover does not change which model a subscription includes. Errors

The new plan names therefore matter alongside the credit. Kimi’s third-party client guide says the new Go tier has no coding quota. Plus and the older Moderato tier can use K3.1 Pro with a context limit of 262,144 tokens. Pro and the older Allegretto tier or above can use up to 1,048,576. A context window is how much text the model can consider at once; a quota reset adds no entitlement to a larger one. Access

Highspeed and extra agents spend the allowance differently

Kimi’s model table puts another tradeoff inside the model selector. K3.1 Pro Highspeed offers approximately two to three times the regular model’s speed at about twice the quota usage. The company describes its capabilities as identical. Selecting it exchanges more allowance for a faster response, rather than increasing the amount of work included in a subscription. Models

Kimi also requires thinking mode for K3.1 Pro. Its model guide recommends starting a new session when moving from older models, saying that switching inside an existing conversation can reduce cache hits and cause unexpected behavior. A cache reuses processing of unchanged earlier text. Those published recommendations identify settings that can affect consumption; they supply no before-and-after measurement of Sunday’s corrections. Models

Subsidiary agents introduce a separate source of activity. Kimi Code’s documentation describes a main agent that can delegate focused assignments, including exploring a codebase, reviewing different implementations and planning a change. Each subsidiary agent works in an independent context and returns a result to the main agent. The user can see one overall assignment while the program runs several model conversations to complete it. Agents

Those agents each consume model tokens independently, Kimi says. Its guide recommends letting the main agent handle simple tasks more economically. Isolation can keep exploratory detail out of the main conversation, and parallel work can shorten a complex job, but neither makes those additional model calls free. The announced reduction in unnecessary delegation targets a cost that the software’s own planning decisions can create. Agents

Sunday’s notice gives no new burst limit or measured reduction in model calls. Kimi

Extra Usage charges a separate balance

Kimi Code membership and the Kimi Open Platform are also billed differently. The company’s FAQ identifies subscription access with included quota for Kimi Code and pay-as-you-go billing for the Open Platform. They use different service addresses and keys. A monthly subscription reset should therefore be read as a membership adjustment, rather than a price reduction for every developer request made to a Kimi model. Guide

Within membership, Extra Usage provides another way to continue after an allowance runs out. The terms describe a prepaid balance, displayed in Chinese renminbi, deducted after applicable subscription and other credits. Charges vary by model and task complexity. This creates an additional payment beyond the subscription fee when the user enables and funds it. Terms

The Extra Usage rules allow a monthly spending limit and say the prepaid balance does not expire. The monthly cap is off by default under the published terms; users have to enable it. The rules generally exclude refunds, transfers and cash redemption. Kimi’s English membership guide says reaching the balance limit guarantees completion of the last model call, rather than an entire multistep task. A job can require further calls after the particular response that used the remaining credit. Terms Plans

Kimi Code shares membership quota across its command-line program, desktop application, editor extension and supported third-party tools. Its account console shows remaining quota and rate-limit status. Running simultaneous sessions on several devices consequently draws on a shared allowance, rather than creating a separate monthly pool for each screen or connection. Plans

The command-line program’s documented /usage command shows token usage, context consumption and quota information. Its /status command identifies the active model and runtime state. Those displays let a user distinguish the model selected from the amount already consumed. The subscription page supplies the monthly account view; a stalled request’s error message supplies the reason that particular call stopped. Commands

Kimi said the adjustment would appear by the end of October 25, using PST. That deadline has passed; subscribers can check their quota display for the credited balance. Kimi