Economics.md
The economics of a usable video
The number that matters is cost per approved deliverable, not cost per generated second.
A simple production model is:
total cost = generation + input/reference charges + retries + postproduction + review labor + storage/delivery
cost per approved second = total cost / seconds accepted for the final deliverable
Neither formula assumes that each generated second becomes useful footage. Keep the rejection rate visible.
A pricing snapshot
Runway's developer pricing page lists credits at $0.01 each. On September 26, 2026, selected listed rates translate as follows. These are prices through that platform, not necessarily direct vendor prices or consumer subscription economics. Runway developer pricing.
| Endpoint / configuration | Listed credits per output second | Derived USD per output second |
|---|---|---|
| Gen-4 Turbo | 5 | $0.05 |
| Gen-4.5 | 12 | $0.12 |
| Veo 3.1 Fast with audio | 15 | $0.15 |
| Veo 3.1 with audio | 40 | $0.40 |
Rates can change; verify the selected endpoint before budgeting. Taxes, extra output options and input charges are not included in this table.
Illustrative sensitivity analysis — not measured performance
Suppose a 10-second candidate costs $1.20 to generate. If one in four candidates is approved in full, the expected generation spend is $4.80 for 10 approved seconds. If approval improves to one in two, it becomes $2.40. If the reviewer spends five minutes per candidate at a hypothetical $60/hour labor cost, review adds $20 in the first scenario and $10 in the second.
These are assumptions to demonstrate the calculation, not claims about Runway or any model's rejection rate. They show why a workflow improvement that reduces review and retries can matter more than a small discount on inference.
| Assumed acceptance rate | Expected candidates per approval | Generation spend at $1.20/candidate | Review at $5/candidate |
|---|---|---|---|
| 25% | 4 | $4.80 | $20.00 |
| 50% | 2 | $2.40 | $10.00 |
| 80% | 1.25 | $1.50 | $6.25 |
A real team may accept only part of a clip, combine several candidates, or reject correlated failures repeatedly. Use observed production data instead of the simple independent-trial assumption once it exists.
Subscription traps for an application builder
Unlimited consumer access is not a guaranteed wholesale API cost. A profitable application must model the contract and endpoint it actually uses. Separate the supplier's list price, negotiated price, concurrency limits and billing treatment of failed requests.
Track usage by project and customer. An apparently healthy average can conceal a customer whose requested style needs many more retries. Treat rush turnaround, higher resolution, reference-heavy editing and human review as distinct cost drivers. Do not subsidize them invisibly inside a flat plan until their usage distribution is understood.
Metrics worth instrumenting
Record attempts per accepted shot, accepted duration, human minutes per deliverable, end-to-end latency including queues, supplier failures, and customer-requested revisions after delivery. Add the final destination and whether the customer reused the workflow next month.
For margin, use collected revenue minus actual variable production and support costs. Model costs alone are an incomplete cost-of-goods estimate. For product value, compare the full previous workflow against the new one, including the steps the user still performs outside the app.
Open with JavaScript for the full viewer.