AI video API costs: 16 model rates and a retry budget
Compare a dated snapshot of 16 AI video API rates in USD, calculate 30-second output costs and see how retries change the budget.
👁 115 reads
Historical price comparison. The source table is dated August 8, 2026, with a September 18 note about renamed or discontinued products and links added later. It is not a live October price feed. This English adaptation keeps the listed USD unit rates and recalculates 30 seconds of output in dollars; check the linked provider before budgeting.
The rates in this snapshot range from $0.04 to $0.40 per generated second, a tenfold difference. Credit bundles, generation allowances and subscription conditions can make comparisons difficult, so the table normalizes 16 model entries to dollars per second and the equivalent cost of 30 seconds. These are listed API rates from a historical source snapshot, not guaranteed current quotes, and 30 seconds of output may require several clips. Taxes, retries, provider markups and output settings can affect the bill. For a separate explanation of Luma credits and API billing, see Luma AI. Guides added for newer models include FLUX 3 Video with a cited starting rate of $0.085 per second, SkyReels V4 with a cited starting rate of $0.08 per second, and Utopai X which is available through a PAI subscription rather than a public API. Check each guide’s date and settings before comparing those figures.
Price table: $1.20 to $12 for 30 seconds of output
The source attributes the following rates to public OpenRouter model pages and labels the table August 8, 2026. Later annotations flag discontinued products. The right-hand column is calculated as the listed per-second rate multiplied by 30; it does not mean every model supports a single 30-second generation. Archived entries are retained for historical context, not presented as available purchase options.
| Model | Provider | USD per second | USD for 30 seconds of output |
|---|---|---|---|
| Wan 2.6 | Alibaba | 0.04 | $1.20 |
| Seedance 2.0 Fast | ByteDance | 0.0404 | $1.212 |
| Veo 3.1 Lite | 0.05 | $1.50 | |
| Seedance 2.0 | ByteDance | 0.0673 | $2.019 |
| Grok Imagine Video 1.5 | xAI | 0.08 | $2.40 |
| Hailuo 2.3 | MiniMax | 0.0817 | $2.451 |
| HappyHorse 1.1 | Alibaba | 0.0988 | $2.964 |
| Wan 2.7 | Alibaba | 0.10 | $3.00 |
| Veo 3.1 Fast | 0.10 | $3.00 | |
| Seedance 2.5 | ByteDance | 0.1028 | $3.084 |
| Kling Video O1 | Kuaishou | 0.112 | $3.36 |
| Kling 3.0 Standard | Kuaishou | 0.126 | $3.78 |
| Hailuo 3 | MiniMax | 0.13 | $3.90 |
| Kling 3.0 Pro | Kuaishou | 0.168 | $5.04 |
| Sora 2 Pro (archived source entry; see status and alternatives) | OpenAI | 0.30 | $9.00 |
| Veo 3.1 | 0.40 | $12.00 |
For the site’s separate price-tracking resource, see the AI video observatory. Check the date shown there before treating a figure as current. This article does not verify that every external or localized price table is refreshed automatically.
What the numbers show
First, the least expensive listed options make small experiments comparatively inexpensive. At the snapshot rates, Wan 2.6 costs $1.20 for 30 seconds of generated output and Seedance 2.0 Fast costs $1.212. That is the cost of output time, not necessarily a finished scene: failed or rejected attempts still consume a budget.
Second, a newer model does not always have a higher listed rate. In this table, Seedance 2.5 is $0.1028 per second, below Kling 3.0 Standard at $0.126. The source also describes Seedance’s ability to generate 30 seconds with audio in a single run. The table alone does not establish current feature exclusivity, quality superiority or a provider’s pricing strategy.
Third, the higher-priced entries can materially change a project budget. The listed full Veo 3.1 rate is ten times Wan 2.6’s. The archived Sora 2 Pro row also sits near the expensive end, but it should not be treated as a current offer. A higher price is not proof of better prompt adherence or audio quality; compare results at the settings you will actually use.
Fourth, retries matter as much as the headline rate. If a usable scene takes three or four attempts, the generation cost can be three or four times the one-attempt figure. At the listed Veo 3.1 rate, four attempts producing 30 seconds each cost 4 × 30 × $0.40 = $48. This is an example calculation, not a prediction that every scene needs four attempts.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →When a subscription can cost less than an API
Repeated generation is one reason to compare subscription offers advertised as unlimited. The source discusses them in the Artlist vs Higgsfield comparison. Its historical example quotes an annual-billing Artlist offer at €41.67 per month before tax for unlimited Seedance 2.5. Using the October 12 reference conversion of 1 euro = $1.1181, that is about $46.59 per month, a converted illustration rather than a verified US checkout price. At the table’s API rate of $0.1028 per second, $46.59 buys about 453 seconds, or roughly fifteen 30-second equivalents. A subscription may cost less above that usage only if the offer is still available and its model access, resolution, queue limits, fair-use rules and commercial terms fit the work. Annual billing also creates a larger upfront commitment. For a few clips per month, usage-based API billing may be cheaper; compare current terms before choosing.
API rates and subscription offers can change. Before setting a project budget, confirm the provider, model version, resolution, audio settings, minimum billed duration and retry policy. The observatory can provide another reference if its date is current. Text-model costs are covered separately in How much does AI cost?.
