Runway says eligible legacy Unlimited subscriptions remain unchanged through November 30, 2026, and then move to its credit-based Max plan. The headline allowance is easy to compare. The harder question is whether 9,500 monthly credits cover the footage your team actually approves after retries, extensions, upscales, and rejected drafts.

Compare one accepted-video batch in APOB

This audit converts AI video credits into accepted final seconds, not raw generations. It uses your own three-job ledger, because model choice, brief difficulty, quality bar, and reviewer behavior make universal burn rates misleading. All Runway terms below were checked against the company's help pages on September 9, 2026 and should be rechecked before a buying decision.

Snapshot the workload before the plan changes

Start with three jobs that represent the month: a routine social clip, a high-precision product shot, and a deadline-sensitive client deliverable. Do not select only clean successes. The ledger should reflect the work that consumes capacity.

Job mix

Record deliverable length, aspect ratio, target resolution, required models, number of variants, and whether the job needs lip sync, extension, upscale, or editing. Link each job to its original brief. If you also use APOB's Runway-style video workflow, keep that workflow in a separate column so provider and cost systems are not mixed.

Acceptance rule

Define “accepted” before counting credits. A clip might need accurate product geometry, readable text, a specific character, clean motion, or one usable five-second segment. If the team changes the bar after seeing a render, mark the change; otherwise the denominator moves invisibly.

Deadline pressure

Note the due date, review window, and maximum acceptable queue time. Runway explains its transition as a move from Explore Mode's relaxed-speed generations to on-demand credit use. That difference matters most when revisions arrive close to delivery, not when the calendar is empty.

Seat pattern

List who generates, reviews, and approves. Runway's credit guide says workspace members draw from a shared balance rather than individual allocations. A team therefore needs owner and job labels on every burn, not just a monthly total.

Log every burn, including the invisible retries

Capture the balance before and after each action or export the available usage record. One ledger row equals one submitted operation. Never reconstruct the month from downloaded favorites; deleted and rejected attempts still shape cost.

Generation burn

Record date, job, model, mode, duration, resolution, quoted or observed debit, and result. The Runway guide notes that listed rates do not always equal the total cost of a generation; duration and tool settings matter. Use the actual account debit when available.

Reconcile the ledger against the opening and closing balance at least once per week. A missing row is not safely treated as zero; flag the difference and investigate it. Keep web-app and API spending separate because Runway states those credit systems are distinct. That separation prevents an API invoice from distorting a web-plan capacity decision.

Extension burn

Treat an extension as a separate action. Record which source clip it extends, the added duration, and whether the final edit uses that material. An extension that fixes continuity but never reaches the export belongs in burn, not accepted seconds.

Upscale burn

Log upscales separately from creation. If a job requires a delivery resolution, include only an output that reaches that requirement in accepted seconds. The APOB AI Video Editor can be tested as a post-generation path, but its operations and costs should remain clearly separated from Runway credits.

Rejected attempts

Give each rejection one primary reason: identity, motion, text, timing, composition, sound, policy, or reviewer change. The purpose is to find expensive failure patterns. Do not call a retry “free exploration” merely because the result was not shipped.

Convert credits into accepted final seconds

Raw seconds show how much media was generated. Accepted seconds show how much survived. Calculate both, then connect every approved segment to the ledger rows that produced it.

Gross seconds

Sum the duration of every generated and extended output. Keep still-image and audio actions in credits but not in video seconds. Gross seconds help explain load, but they do not describe deliverable value.

Accepted seconds

Count only unique footage in the approved cut. If two ten-second outputs contribute the same three-second segment, count three accepted seconds, not six. Preserve the edit decision list or timecodes so the total can be audited.

Credits per accepted second

Use total credits consumed by the job ÷ unique approved seconds. Keep the raw credits per video second beside that accepted-output metric. This is an observed project measure, not a promised AI video credit cost. Report the date, models, and acceptance rule next to it. A number without those conditions invites a false comparison.

Retry ratio

Use submitted generation and extension attempts ÷ accepted outputs, or report accepted outputs per attempt. Choose one definition and keep it fixed. Segment the ratio by failure reason; a high product-text retry rate suggests a different workflow decision than a high reviewer-change rate.

Also calculate review minutes per accepted second. Credits describe platform consumption, while review time reveals the human cost of searching, comparing, and repairing. A lower-credit path can still be the worse production choice if specialists spend more time recovering usable footage. Keep the measures separate so neither is used as a proxy for the other.

For a matched comparison, send the same brief and acceptance rule through the APOB AI Video Generator. Compare approved outputs, elapsed review time, and repair actions—not incompatible credit labels.

Stress-test rollover and shared-workspace risk

Runway's current documentation states that Max includes 9,500 monthly credits and allows up to one month of unused credits to roll into the following month. It also says credits are spent in the order current-month plan credits, rollover credits, then purchased credits. Test how that policy interacts with your billing date.

Quiet-month balance

Take a low-volume month from the ledger and project the unused plan balance. Do not assume rollover compounds indefinitely; the documented allowance is up to one month. Mark the exact billing date because that boundary controls expiry and replenishment.

Launch spike

Build a launch-week scenario from a real deliverable list. Apply the observed credits per accepted second and retry mix, then add a clearly labeled contingency. The result is a planning scenario, not a prediction of model behavior.

Spending order

Separate monthly, rollover, and purchased balances. Runway says purchased credits do not expire, while monthly credits have plan-specific rules. A single total balance can conceal which pool will disappear first.

Shared pool

Assign a project cap and approver. A workspace-wide balance means one heavy job can consume another team's launch reserve. Review usage at a fixed daily time during peaks, and require a job code before expensive generations.

The official transition notice also distinguishes legacy Unlimited's 2,250 credits plus Explore Mode from Max's 9,500 credits and no Explore Mode. Do not describe Max as unlimited.

Set a stop-loss rule before November 30

A stop-loss turns the ledger into an operational decision. It should trigger early enough to test another path without jeopardizing delivery.

Trigger metric

Choose one metric your team can read daily: credits per accepted second, rejected attempts per accepted clip, remaining credits per scheduled deliverable, or projected balance at billing. Set the threshold from your baseline rather than a public average.

Escalation owner

Name the person who can pause generations, change the brief, approve more credits, or reroute the job. Without an owner, the threshold becomes a dashboard decoration.

Give that owner a one-page weekly report: opening balance, consumed credits, accepted seconds, retry ratio, projected billing-date balance, scheduled rush work, and top rejection code. A stop-loss should start a conversation early enough to change a workflow, not merely explain an exhausted balance after delivery has already slipped.

Record the decision and its date each time the threshold fires. Later, compare whether the intervention reduced rejection burn or merely shifted cost into review and editing. A stop-loss earns its place only when it changes an observable production outcome.

Alternative path

Define what happens next: simplify the shot, move text to post-production, choose another model, or run the matched brief through APOB's multi-model workflow. Keep the same acceptance rule so the comparison remains useful.

Review date

Review this audit before November 30, 2026 and again on October 9 or whenever Runway changes plan terms. Confirm the account's actual migration message; the help article says eligible subscribers were emailed details, and annual subscribers have a different choice path from monthly users.

The decision is not “9,500 sounds larger than 2,250.” It is whether the Runway Max plan covers the accepted footage, retry behavior, collaboration pattern, and deadline risk your own ledger reveals.

Sources

• Runway — Unlimited Plan Is Switching to Max

• Runway — How Do Credits Work?

• APOB AI — Runway AI Video Generation

• APOB AI — AI Video Generator

• APOB AI — AI Video Editor