How coding plans count usage: credits, requests, dollars, and windows

The credit unit and every window figure in the Z.ai rows. Verbatim: "Model credit usage = (Input tokens × Input multiplier + Cached Input tokens × Cached Input multiplier + Output tokens × Output multiplier) / 10,000"; the multiplier table (GLM-5.3 input 6.9, cached input 1.7, output 24; GLM-5.3-Flash 2.3, 0.56, 8), which is where the fourteen-times ratio between an output token and a cached input token is read; "Each plan is subject to both a 5-hour usage limit and a weekly usage limit" with Lite 2,000 / 10,000, Pro 12,000 / 60,000 and Max 28,000 / 140,000 credits; "5-hour credits: Dynamically refreshed; credit quota resets 5 hours after consumption"; "Weekly credits: Activated upon subscription; resets every 7 days"; and "During off-peak hours, model usage is charged at 50% of the standard credit rate. Peak hours: Monday to Friday, 14:00–18:00 Singapore Standard Time (UTC+8)." On 2026-09-29 the page also read "From September 25 to October 7, 2026, all-day usage will be charged at the off-peak rate", which the Z.ai note quotes with its dates.

The monthly prices in the Z.ai row — Lite $18, Pro $80, Max $168 — read in a browser because the page is JavaScript-rendered; on 2026-09-29 it opened on yearly billing at 30% off, showing $12.6, $56 and $117.6 a month beside those monthly prices. Subscribe controls were live on both reads, which is the basis for recording the plan as open to new subscribers. The same page carries the concurrency FAQ answer that rate limits are tied to the plan tier "with the general principle being Max > Pro > Lite", with no numeric figure anywhere.

The CNY lineup row: Andante ¥49, Moderato ¥99, Allegretto ¥199 and Allegro ¥699 per month, with sizing of about 30, 60, 150 and 360 Agent uses. The unit wording is verbatim: "All Kimi membership features share one credit pool and are calculated based on actual token consumption"; "You can view your current credit balance as a percentage"; and the shared-pool consequence that if one feature uses up the credits, other features are affected. This page also carries the ambiguous sentence quoted in the unsettled section: "Kimi Code also has a separate limit of 5 hours per week, which applies only to Kimi Code and does not affect other membership features." No signup-availability statement appears on it, which is what that cell records. The page read the same on 2026-09-29, while Kimi's developer documentation now calls this lineup the legacy plans that existing members keep, which is what the signup cell records.

The row for the plans released in September, read in a browser on 2026-09-29: Plus $19, Pro $39, Max $99 and Ultra $199 per month, with annual totals of $180, $372, $948 and $1,908, and relative allowance labels rather than numbers ("2 倍", "5 倍", "10 倍 Agent 额度"). Every tier's call to action reads "订阅" (subscribe) under a banner that the new membership plans are now available, which is the basis for the open cell. On 2026-09-12 the same prices carried the names Moderato, Allegretto, Allegro and Vivace and a "预约订阅" (reserve a subscription) button.

The windows rows for Kimi Code and the developer-documentation half of the contradiction. Verbatim on 2026-09-29: "The new Kimi membership plans are now available, with pricing unchanged"; "For new members, the weekly quota limit is removed, and only the rolling 5-hour rate window remains"; "Existing subscribers on legacy plans are not affected: the tier name, quota rules, and auto-renewal all remain as they are"; and the 5-hour window as a rate limit: "too many requests in a short time trigger rate limiting, which recovers automatically once the window rolls over". Both groups share the membership's monthly total: "if the monthly total is reached, Kimi Code quota is frozen until the monthly quota resets or you upgrade." On 2026-09-12 the page described one 7-day quota and the 5-hour window, with no new or legacy split.

The model multiplier quoted in the section on why the unit matters: "k3 (1M) consumes about twice as much quota as k3-256k". The same page gates models by tier (k3 from Moderato or Plus up, the up-to-1M context window from Allegretto or Pro up), which is why a tier's price does not by itself say which models it reaches.

The MiniMax row: Plus $22, Max $55 and Ultra $132 per month, each listing "Quota windows: 5-hour rolling and weekly windows" and an "Agent usage" sizing of 3-4, 4-5 and 6-7 agents. The unit wording is verbatim: "Token Plan subscriptions provide a monthly usage quota and access to eligible resources through the Subscription Key. Usage is shown as a usage bar in the console." A live subscribe control was present, which is the basis for recording signup as open. No numeric allowance is published anywhere on the page.

The consumption rule and the monthly cell in the windows table: usage "deducts from the included Token Plan quota according to the corresponding endpoint pricing", and "unused subscription quota does not carry over to the next billing cycle". No reset behaviour is stated for the weekly or monthly window on the pages read, which is what that cell records rather than inferring one.

The single source for both Alibaba Coding Plan rows; the page stated "Updated at: 2026-09-11" on the first read, and the figures below read the same on 2026-09-29. Verbatim: price "¥ 200/month"; the unit definition "Each query consumes quota based on the number of model calls. Simple tasks typically use 5–10 calls, while complex tasks may use 10–30 or more."; the quotas "Up to 6,000 requests per 5 hours / Up to 45,000 requests per week / Up to 90,000 requests per month"; the stacking rule "The three limits above apply simultaneously as caps, and whichever is reached first takes effect" together with "The limits are not cumulative and do not represent a guaranteed capacity distributed evenly over time"; the resets ("Usage from 5 hours ago is restored", "Resets every Monday at 00:00:00 (UTC+08:00)", and a monthly reset on the subscription renewal date); the throttling clause "When overall platform load is high, or when a single account consumes an unusually concentrated amount of resources within a short period, we may temporarily queue, throttle, or briefly interrupt requests from that account"; the rationing sentence "Slots are limited and available on a first-come, first-served basis. New slots are restocked daily at 09:30:00 (UTC+08:00)" with the Lite tier closed to new subscriptions as of March 2026; and the withheld concurrency thresholds, which are "part of platform security and stability policies".

The Token Plan rows, re-read on 2026-09-29: Lite, Essential, Standard and Pro at "Original price 60 CNY/month Limited-time 39 CNY/month", 120 and 79, 180 and 139, and 600 and 499 CNY, with monthly quotas of 11,500, 25,500, 45,000 and 180,000 Credits and 1–2, 2–3, 3–4 and 6–8 concurrent agents. The weekly quota ended on September 22, 2026 ("When the weekly quota was removed on September 22, 2026, the remaining quota … was reset once"), and "A 30-day subscription cycle starts from the subscription date"; on 2026-09-12 the page gave 2,500, 10,000 and 40,000 Credits per 7-day window timed from first use, and no Essential tier. Unused Credits do not carry forward, and no 5-hour window or peak-time policy appears. This is a general Model Studio subscription rather than a coding-specific plan; it appears here because the vendor's coding tooling documents it as a billing option.

The OpenCode Go rows; the page states it was last updated 2026-09-28 (2026-09-10 on the first read). Verbatim in English: "Usage limits are defined as monthly dollar amounts."; "Each model has the following usage limits: 5-hour — 20% of the monthly limit; weekly — 50%; and monthly — 100%."; and the DeepSeek pass-through "Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday; all other hours, including weekends, are Off-Peak", where peak token rates are double the off-peak rates. The per-model monthly limits quoted (Kimi K3 $15, GLM-5.3 $15 and GLM-5.3-Flash $60 on Go; $60, $120 and $180 on Go Plus, a $40 tier added since the first read with the same token prices) come from the same table, which budgets each model separately rather than pooling them. The page states that the limits may change based on early usage and feedback, and it publishes no reset behaviour for the three windows. It was served in Korean localisation on first read; the English sentences were retrieved separately from the same URL.

The $10 per month price in USD and the live subscribe control that is the basis for recording signup as open. The $40 Go Plus tier was read on the documentation page on 2026-09-29.

The DeepSeek row, read in a browser. Confirms there is no subscription of any kind and that billing is balance deduction: "The expense = number of tokens × price. The corresponding fees will be directly deducted from your topped-up balance or granted balance". The peak policy is verbatim: "Off-peak rates are half of the peak rates. Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday, excluding Chinese public holidays. All other hours are off-peak, including weekends and Chinese public holidays in full." The holiday clause was not on the page on 2026-09-12. The same page publishes per-model concurrency limits (2500 for deepseek-flash, 500 for deepseek-v4-pro), which is the basis for the sentence that DeepSeek is the one vendor here printing concurrency figures.

The reference-grammar row, from an article marked "Updated this week" on 2026-09-29 (dated August 8, 2026 on the first read). Verbatim: "Max 5x includes five times the Pro plan's per-session usage allowance"; "Your session-based usage limit will reset every five hours. Max plans also have a weekly usage limit that applies across all models. The weekly limit resets at a fixed time each week that is assigned to your account. Your reset day and time stay the same regardless of when you start using Claude or when your subscription begins"; and the reserved clause "we may limit your usage in other ways, such as weekly and monthly caps or model and feature usage, at our discretion". This article's "Priority access" benefit is early access to new models and features, so the peak sentence cites the Pro article. Prices confirmed at $100 and $200 per month.

The $20 per month price in the same row, from an article marked "Updated this week" on 2026-09-29, and the confirmation that no absolute unit is published for it either: "The Pro plan offers more usage per session than the Free plan" ("at least five times" on the first read), and "The number of messages you can send will vary based on message length, including the length of files you attach, the length of your current conversation, and the model or feature you use", with a weekly limit that "applies across all models". Its benefits list carries the peak sentence: "Priority access to Claude during high-traffic periods".

Our own row. The page publishes three tiers at $50, $100 and $200 per developer per month, described as a smaller allowance with 5-hour and weekly limits, a larger allowance with a weekly limit only, and no 5-hour or weekly limits and "no total usage cap"; the three included models Kimi K3, GLM-5.3 and DeepSeek V4.1 Flash; and the sentence "Peak-time restrictions apply to every plan: requests may slow, queue, or time out. Priority is $200 → $100 → $50." No allowance unit, no allowance amount and no reset rule appears anywhere on the page, which is why those cells say the terms are confirmed before activation.

The limits table in our documentation repeats the three tiers and adds "Peak-time restrictions apply to every plan, including $200. Requests may slow, queue, or time out. Priority is $200 → $100 → $50. An uncapped usage allowance does not guarantee processing speed or concurrency." It then states the condition this page reports rather than filling in: "Exact allowances and request and concurrency limits are confirmed before activation."

The Coding Plan section of our pricing page, which carries the same three tiers and the sentence "Allowance amounts, supported tools, renewal dates, and cancellation terms are confirmed before activation", plus the note that priority "is not a latency or availability guarantee". Its FAQ supplies the signup cell in our row: "Account creation, product access, and plan activation are separate steps."