AI video generation API pricing is harder to compare than text or image pricing because the billing unit changes by provider. Google Veo and OpenAI Sora expose per-second video prices, BytePlus Seedance examples are tied to token/resource-pack consumption, and every route has extra operational questions around duration, resolution, async jobs, retry behavior, and model lifecycle.
This comparison was checked on June 17, 2026 Asia/Shanghai using official OpenAI pricing, OpenAI deprecations, Google Gemini API pricing, Google Veo documentation, and BytePlus ModelArk Seedance pricing sources plus a live Flatkey pricing-page snapshot. Treat the prices below as point-in-time planning evidence. Before production traffic, verify the current provider page, the exact Flatkey row in Flatkey pricing, the endpoint family, availability, quota behavior, and a small test call.
Quick Answer: How To Compare AI Video Generation API Pricing
For AI video generation API pricing, compare these fields before looking for the lowest headline number:
- Billing unit: per second, token/resource-pack consumption, per generated video, or a routed dashboard unit.
- Duration: 4 seconds, 5 seconds, 8 seconds, 15 seconds, 20 seconds, or provider-specific supported ranges.
- Resolution: 480p, 720p, 1080p, 1024p, 4K, landscape, portrait, or adaptive output.
- Completion policy: whether failed jobs are charged, whether retry jobs are new billable jobs, and how async status is logged.
- Lifecycle risk: preview status, deprecation date, shutdown date, and whether a replacement is listed.
- Route proof: exact Flatkey model row, endpoint type, availability status, usage logs, and billing label.
For AI video generation API pricing, the short version is this: Veo 3.1 is easiest to normalize by seconds and resolution, Sora 2 has clear per-second prices but a September 24, 2026 shutdown date, and Seedance 2.0 needs token/resource-pack checks instead of a simple per-second shortcut.
AI Video Generation API Pricing Comparison Table
The table below gives a normalized starting point for AI video generation API pricing. It does not rank the models as cheapest or best because these providers are not selling identical units.
| Model Family | Official Billing Shape Checked | Example Public Price Points | Operational Risk | Flatkey Check |
|---|---|---|---|---|
| BytePlus Seedance 2.0 | Token/resource-pack based examples; actual usage is tied to returned completion_tokens. |
For input without video, 5-second 16:9 examples list Seedance 2.0 at $0.35 for 480p, $0.76 for 720p, and $1.87 for 1080p. | Do not convert directly to per-second pricing without token and duration context. | Check bytedance/seedance-2.0 and bytedance/seedance-2.0-fast, endpoint type, group, status, and usage logs. |
| Google Veo 3.1 | Paid-tier per-second pricing by variant and resolution. | Standard is $0.40/sec at 720p or 1080p and $0.60/sec at 4K. Fast is $0.10/sec at 720p, $0.12/sec at 1080p, and $0.30/sec at 4K. Lite is $0.05/sec at 720p and $0.08/sec at 1080p. | Preview models may change; older Veo 3 and Veo 2 rows are scheduled for June 30, 2026 shutdown. | Check veo-3.1-generate-preview, veo-3.1-fast-generate-preview, and veo-3.1-lite-generate-preview. |
| OpenAI Sora 2 | Per-second pricing by model, resolution, and standard versus batch path. | sora-2 720p is $0.10/sec standard and $0.05/sec batch. sora-2-pro ranges from $0.30/sec to $0.70/sec standard, depending on resolution. |
OpenAI lists Videos API, sora-2, and sora-2-pro for shutdown on September 24, 2026 with no replacement listed. |
Check sora-2 and sora-2-pro, but treat them as short-horizon rows unless the lifecycle status changes. |
Seedance Pricing: Token And Resource-Pack Math
BytePlus ModelArk's Seedance material should be read differently from a simple per-second price sheet. The pricing source says actual consumption is subject to the usage.completion_tokens or completion_tokens value returned after the API call. That makes Seedance a poor fit for copy-pasting into a static AI video generation API pricing table without the request shape.
For a 5-second 16:9 input-without-video example, the saved BytePlus pricing page listed these estimates:
| Seedance Model | 480p | 720p | 1080p | Planning Note |
|---|---|---|---|---|
| Seedance 2.0 | $0.35 | $0.76 | $1.87 | Use as an example for one request shape, not a universal per-second rate. |
| Dreamina Seedance 2.0 Fast | $0.28 | $0.60 | Not supported | Fast can lower the example cost, but check supported resolution and quality tradeoffs. |
| Dreamina Seedance 2.0 Mini | $0.18 | $0.38 | Not supported | Useful for budget planning when lower-resolution output is acceptable. |
The Seedance API reference checked for this article describes a video generation API for models such as Seedance 2.0 and includes model IDs such as dreamina-seedance-2-0-260128. It also shows request controls that affect the final bill, including resolution, duration, reference media, and output options. For procurement, ask finance to approve a token/resource-pack budget rather than a one-line per-second estimate.
Veo 3.1 Pricing: Per Second, But Resolution Changes The Budget
Google's Gemini API pricing page is easier to normalize for video because Veo 3.1 is listed per second. The current paid-tier table checked for this article includes:
| Veo 3.1 Variant | 720p | 1080p | 4K | Notes |
|---|---|---|---|---|
| Veo 3.1 Standard | $0.40/sec | $0.40/sec | $0.60/sec | Default video with audio price. |
| Veo 3.1 Fast | $0.10/sec | $0.12/sec | $0.30/sec | Lower-cost fast variant; verify output requirements. |
| Veo 3.1 Lite | $0.05/sec | $0.08/sec | Not supported | Good comparison row for lower-resolution budgets. |
The same pricing page says customers are charged only if a video is successfully generated when an audio processing issue prevents generation. That matters for AI video generation API pricing because a failed-job policy can change the true cost of a creative workflow with many retries.
The Veo 3.1 guide also shows that video generation uses long-running operations: submit a job, poll the operation until it is done, then download the generated video. Operationally, this means your cost dashboard should log job ID, request parameters, completion status, retry count, and final seconds. A plain provider price table will not tell you whether your application is spending most of its budget on high-resolution final renders or repeated drafts.
veo-3.0-generate-001, veo-3.0-fast-generate-001, or veo-2.0-generate-001, update the route plan before launch.Sora Pricing: Per Second With A Shutdown Clock
OpenAI's Sora pricing is straightforward at the unit level and risky at the lifecycle level. The official OpenAI pricing table currently lists video generation models as prices per second. The docs search result checked for this article returned these Sora 2 rows:
| Sora Model | Resolution | Standard Price | Batch Price | Planning Note |
|---|---|---|---|---|
sora-2 |
720p | $0.10/sec | $0.05/sec | Lowest listed Sora row, but still subject to shutdown risk. |
sora-2-pro |
720p | $0.30/sec | $0.15/sec | Pro model at 720p. |
sora-2-pro |
1024p | $0.50/sec | $0.25/sec | Higher-resolution planning row. |
sora-2-pro |
1080p | $0.70/sec | $0.35/sec | Highest Sora row checked in the table. |
sora-2, sora-2-pro, and related snapshots are scheduled for API removal on September 24, 2026, with no recommended replacement listed. Do not build a long-lived production plan on Sora 2 without a migration path.This is where a normal AI video generation API pricing comparison can mislead teams. A $0.10/sec row may look attractive for a short experiment, but a model with a fixed shutdown date needs a separate migration budget. If you use Sora 2 before September 24, 2026, log the exact model ID, output duration, resolution, batch or standard path, and replacement decision in the same dashboard.
Flatkey Catalog Snapshot For Video Models
Flatkey's public pricing page was checked on June 17, 2026 Asia/Shanghai. The server-returned pricing snapshot contained 638 model rows, top-level pricing version a42d372ccf0b5dd13ecf71203521f9d2, and endpoint families including gemini, openai, openai-video, openai-response, and image-generation. Relevant video rows were present, but every row in this issue's target set showed unknown_failure availability status.
| Flatkey Row Checked | Endpoint Types | Pricing Fields Seen | Status In Snapshot | Action Before Production |
|---|---|---|---|---|
sora-2, sora-2-pro |
openai |
model_price: 0.3 and model_price: 0.5, quota_type: 1 |
unknown_failure |
Check lifecycle status and do not assume route readiness beyond the September 24, 2026 deprecation window. |
veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview |
gemini, openai |
Standard and Fast rows showed model_price: 0.4 and model_price: 0.15; Lite showed ratio fields. |
unknown_failure |
Confirm current row, endpoint family, quota type, and success logs before any routed workload. |
bytedance/seedance-2.0, bytedance/seedance-2.0-fast |
openai-video, openai |
model_ratio: 0.391, completion_ratio: 4.956521739130435 |
unknown_failure |
Map Flatkey units to returned usage fields before quoting a customer-facing price. |
This Flatkey evidence is useful for route planning, but it is not a support guarantee. Use the catalog as the first stop: find the exact row, confirm endpoint type, review the status, set a quota, run a small smoke test, and compare the usage log against your estimate.
A Normalized Cost Formula For Video Generation
Use a formula like this before comparing Seedance, Veo, Sora, or any routed model:
Estimated video workflow cost =
successful generated seconds at requested resolution
+ token or resource-pack usage from returned usage fields
+ reference image, video, or audio input cost
+ preview, draft, and final-render attempts
+ retry and failed-job billing policy
+ route-specific Flatkey catalog unit and usage-log result
+ lifecycle migration cost for deprecated or preview models
For Veo and Sora, the generated-seconds line is usually the easiest place to start. For Seedance, the returned token usage and pricing calculator matter more. For Flatkey, the important proof is not a spreadsheet cell; it is the combination of current model row, endpoint family, quota type, availability, and the post-call usage log.
Procurement Checklist For Video Generation APIs
Teams comparing AI video generation API pricing should collect these fields before approving spend:
| Field | Why Finance Cares | Why Engineering Cares |
|---|---|---|
| Exact model ID | Prevents budgeting against an alias, preview row, or deprecated model. | Prevents calls to the wrong endpoint or unsupported request shape. |
| Billing unit | Separates per-second, per-token, resource-pack, and route-specific billing. | Determines which usage field must be logged. |
| Duration and resolution | Shows why 4K final renders cost more than 720p drafts. | Controls request parameters and output validation. |
| Async job status | Explains whether failed or pending jobs affect spend. | Requires polling, retry, timeout, and webhook handling. |
| Lifecycle date | Creates a migration reserve for deprecated models. | Sets replacement work before shutdown. |
| Flatkey row and usage log | Confirms the billable route used by the team. | Confirms endpoint type, status, quota, and request outcome. |
Recommended Decision Path
If your team needs a long-lived production route, start with models that do not have an imminent shutdown date, then verify the exact Flatkey row and provider route. If you need a short-lived Sora 2 experiment before September 24, 2026, document the replacement path before shipping it to users. If Seedance looks attractive for your workflow, estimate from token/resource-pack examples and validate against returned usage. If Veo 3.1 is the candidate, separate standard, fast, and lite variants by resolution and generation quality.
Flatkey's role in this workflow is operational visibility. One pricing page should not be your only control surface. Use Flatkey pricing to find the row, then use quotas, usage logs, and endpoint checks to keep AI video generation API pricing visible after the first test call.
FAQ
What is the best way to compare AI video generation API pricing?
Normalize the billing unit first. Record whether each provider bills per second, per token/resource-pack, or through another routed unit, then add duration, resolution, success or failure billing, retries, and lifecycle risk.
Is Veo 3.1 priced per second?
Google's Gemini API pricing page checked for this article lists Veo 3.1 paid-tier video prices per second by variant and resolution. Standard, Fast, and Lite have different 720p, 1080p, and 4K support.
Is Sora 2 safe for a long-lived production video route?
Not without a migration plan. OpenAI's deprecations page lists the Videos API, sora-2, and sora-2-pro for shutdown on September 24, 2026, with no replacement listed in that table.
Is Seedance priced per second?
The BytePlus materials checked for this article use token/resource-pack style examples and say actual consumption is tied to returned usage fields such as completion_tokens. Treat Seedance examples as request-shape estimates, not universal per-second rates.
How should I check video model pricing in Flatkey?
Search the exact model row in Flatkey pricing, confirm endpoint type, group, quota type, availability status, and current pricing label, then run a small smoke test and compare the usage log with your estimate.
For the broader cost stack, compare this guide with the AI model pricing comparison, the AI image generation API pricing comparison, the Seedance API access guide, and the Sora 2 API deprecation watch.
View Pricing: Use the current Flatkey pricing catalog to verify model rows, endpoint families, status, and usage controls before routing production video workloads.
