BuilderBuilder
← Capacity index
Subscription Capacity Index · Provisional v0.2Provisional

$39 plan. About $78 of weekly API-priced Kimi capacity.

Measured from a real Kimi K3 coding sweep. Workload-specific, provisional, and not Kimi's internal serving cost.

Plan price
$39/mo
Weekly estimate
$78.29
Monthly-cycle estimate
$372.10
Status
Provisional
Primary receipt

One workload. Weekly and monthly quota receipts.

The first-party post-run receipt corroborates weekly use at 51.17% and reports 10.73% monthly-cycle use. No paired pre-run receipt was preserved; both capacity figures are linear, workload-specific extrapolations.

Plan price
$39/month
Weekly quota consumed
51%
Recorded API-equivalent usage
$39.93
Estimated full weekly capacity
$78.29
Rounding interval
$77.53–$79.06
First-party 7-day meter
51.17%
Monthly-cycle meter consumed
10.73%
Estimated full monthly-cycle capacity
$372.10
Monthly rounding interval
$371.93–$372.28
Why price matters

Capacity changes the work you can attempt.

Price buys capacity. Capacity determines how many serious attempts, expensive reasoning modes and completed builds fit before the limit.

Price does not affect RemakeBench's quality score.Cost and capacity stay separate from output quality.

Token and pricing receipt

The arithmetic is the evidence.

Official Kimi K3 API rates verified 2026-07-20. The values below are API-equivalent retail value, not a subscription invoice or provider serving cost.

Kimi K3 recorded token usage and API-equivalent value
Token classRecorded tokensRate / 1MCalculationValue
Fresh input2,024,673$3.002,024,673 × $3.00 / 1,000,000 = $6.074019$6.074019
Cache read73,482,718$0.3073,482,718 × $0.30 / 1,000,000 = $22.0448154$22.0448154
Cache creation0$3.000 × $3.00 / 1,000,000 = $0$0
Output787,197$15.00787,197 × $15.00 / 1,000,000 = $11.807955$11.807955
Observed API-equivalent value$39.9267894$39.93 displayed
Total processed76,294,588 tokensFresh input + cache reads + cache creation + output
Open the official Kimi K3 pricing source
Workload accounting

Successful builds are only part of the bill.

Failed attempts, setup probes and receipt extraction still count against a real subscription quota.

Completed benchmark subset9 tasks≥$27.82 API-equivalent lower bound

Nine completed Kimi Code benchmark tasks. One MacBook request has no reconciled usage record.

All recorded window activity$39.93$12.11 beyond completed tasks

Reconciled attempts, probes and receipt-generation work consumed quota even when they did not become a successful deliverable.

Capacity history

A time series starts with one honest point.

Future providers, plans, models, weekly snapshots and verified allowance adjustments can join this timeline without rewriting it.

  1. Snapshot 001Provisional
    Provider
    Kimi / Moonshot AI
    Plan
    Allegretto · $39/month
    Model / mode
    Kimi K3 · max
    Client
    Kimi Code CLI
    Quota window
    Weekly · reset shown in 3d 11h 37m
    Original weekly reading
    51% used
    First-party 7-day receipt
    51.17% used
    Observed value
    $39.93
    Estimated full capacity
    $78.29
    Confidence / status
    Provisional
    Monthly-cycle meter
    10.73% used · resets 2026-08-17; timezone not displayed
    Estimated full monthly-cycle capacity
    $372.10 · $371.93–$372.28
    Paired pre-run receipt
    Not preserved
    Base allowance
    Allegretto plan allowance
    Promotional / reset adjustments
    None recorded
Methodology and caveats

What this number does—and does not—mean.

This is one provisional observation of one account, plan, workload and weekly window. Future dedicated-account measurements should improve confidence.

  1. 01

    usage.record events were counted; duplicate step.end.usage records were not summed again.

  2. 02

    API list price measures retail-equivalent value. Provider internal serving cost remains unknown.

  3. 03

    The original whole-number weekly reading introduces a 50.5%–51.5% interval around 51%. The later first-party receipt corroborates it at 51.17%.

  4. 04

    The weekly and monthly estimates assume quota use and API-equivalent value scale linearly through their respective windows.

  5. 05

    This workload was extremely cache-heavy. Other workload mixes can produce a different API-equivalent ceiling.

  6. 06

    No paired pre-run monthly-cycle receipt was preserved. The first-party receipt is post-run and does not display a provider capture time or reset timezone, so attribution still relies on reset timing and the operator's attestation that all relevant window usage occurred on the measured computer.

  7. 07

    The monthly estimate is calculated directly from the 10.73% monthly-cycle meter. Its interval reflects the meter's two-decimal display precision; it is not derived from the weekly estimate.

  8. 08

    The observation covers one plan and account. It is not a guaranteed or universal raw-token allowance.

Sanitized public evidence

Inspect the versioned receipt.

The immutable provisional bundle publishes totals, formulas, source dates, private-event hashes and a metadata-stripped copy of the first-party quota receipt. It excludes raw prompts, reasoning, local paths, credentials and session identifiers.

/evidence/subscription-capacity-index/kimi-allegretto/v2
RemakeBenchCapacity index