TL;DR
Directly comparable first-party AI video API rates range from $0.03 to $0.70 per generated second, or $0.30 to $7.00 per normalized 10-second equivalent. Not every API supports an exact 10-second request.
Veo 3.1 Lite has the lowest listed 720p video-only rate, while Sora 2 Pro at 1080p is the highest configuration in the main comparison. For production, compare cost per usable clip after retries, rejected outputs, review, editing, and storage—not only the price of one generation.
AI Video API Pricing at a Glance
The table below uses current first-party list prices or official credit conversions. The 10-second amount is a comparison unit, not a claim that every endpoint supports a 10-second output.
| Provider route | Priced configuration | Price per second | 10-second equivalent | Duration or status | Main caveat |
|---|---|---|---|---|---|
| Google Veo 3.1 Lite | 720p, video only | $0.03 | $0.30 | 4, 6, or 8 seconds; Preview | Fixed-quota access and regional availability vary |
| Runway Gen-4 Turbo | Standard API generation | $0.05 | $0.50 | 5 credits/second | Runway credits cost $0.01 each |
| Google Veo 3.1 Fast | 720p, video only | $0.08 | $0.80 | 4, 6, or 8 seconds; GA | Audio availability depends on the API surface and endpoint |
| Kling 3.0 | 720p, no native audio | $0.08 | $0.84 | 3–15 seconds | 1080p and audio configurations cost more |
| OpenAI Sora 2 | 720p | $0.10 | $1.00 | Up to 20 seconds; Deprecated | Videos API shuts down September 24, 2026 |
| Kling 3.0 Turbo | 720p, native audio | $0.11 | $1.12 | 3–15 seconds | 1080p is $0.14/second |
| Runway Gen-4.5 | Standard API generation | $0.12 | $1.20 | 12 credits/second | Higher price must produce a measurable quality gain |
| Kling 3.0 | 720p, native audio, no voice control | $0.13 | $1.26 | 3–15 seconds | 1080p is $0.168/second |
| Google Veo 3.1 | 720p or 1080p, video only | $0.20 | $2.00 | 4, 6, or 8 seconds; GA | 4K and audio use separate pricing tiers |
| OpenAI Sora 2 Pro | 720p | $0.30 | $3.00 | Up to 20 seconds; Deprecated | 1024p and 1080p cost more |
| OpenAI Sora 2 Pro | 1080p | $0.70 | $7.00 | Up to 20 seconds; Deprecated | Do not make it a new long-term dependency |
Quick answers
- Lowest listed video-only rate: Veo 3.1 Lite at $0.03 per second for 720p.
- Immediate migration priority: Sora 2, because its API shuts down on September 24, 2026.
- Not directly comparable at one fixed rate: Seedance 2.0, which uses configuration-dependent, token-metered pricing.
Pricing and Availability Note
Prices exclude taxes, storage, transfer, editing, rejected outputs, and human review. Before deployment, verify the exact model ID, platform, region, resolution, audio mode, duration, billing policy, and lifecycle date.
Which AI Video API Is Cheapest in 2026?
For a directly comparable 720p video-only configuration, Veo 3.1 Lite has the lowest current first-party list price at $0.03 per second. Its documented 8-second output has a $0.24 generation baseline.
At 1080p, Veo 3.1 Lite is $0.05 per second, Kling 3.0 starts at $0.112 per second without native audio, and Sora 2 Pro reaches $0.70 per second. Because features and supported durations differ, use list price to shortlist routes—not to declare a production winner.
Cheapest production route =
lowest total cost among routes that pass the workload's acceptance criteria
Confirm the result with your own prompts, acceptance rate, retries, and reviewer time.
AI Video API Pricing by Provider
OpenAI Sora 2 and Sora 2 Pro API Pricing
OpenAI's official video model pricing lists the following prices per output second:
| Model | Resolution | Standard price | Batch price | 10-second standard cost |
|---|---|---|---|---|
| sora-2 | 720p | $0.10/second | $0.05/second | $1.00 |
| sora-2-pro | 720p | $0.30/second | $0.15/second | $3.00 |
| sora-2-pro | 1024p | $0.50/second | $0.25/second | $5.00 |
| sora-2-pro | 1080p | $0.70/second | $0.35/second | $7.00 |
The OpenAI video generation documentation covers asynchronous jobs, synchronized audio, image guidance, editing, extensions, and outputs of up to 20 seconds. Both Sora 2 models and the Videos API are deprecated, so use these rates mainly for existing workloads and migration testing.
Google Veo 3.1 API Pricing: Lite, Fast, Audio, and Resolution Tiers
Google's current generative AI pricing page lists these output-second rates:
| Model | Configuration | 720p | 1080p | 4K |
|---|---|---|---|---|
| Veo 3.1 Lite | Video only | $0.03/s | $0.05/s | Not listed |
| Veo 3.1 Lite | Video + audio | $0.05/s | $0.08/s | Not supported |
| Veo 3.1 Fast | Video only | $0.08/s | $0.10/s | $0.25/s |
| Veo 3.1 Fast | Video + audio | $0.10/s | $0.12/s | $0.30/s |
| Veo 3.1 | Video only | $0.20/s | $0.20/s | $0.40/s |
| Veo 3.1 | Video + audio | $0.40/s | $0.40/s | $0.60/s |
Google documents 4-, 6-, and 8-second outputs. The standard and Fast -001 Agent Platform endpoints are GA with retirement dates of November 17, 2026 or later; Veo 3.1 Lite is Preview.
Audio caveat: Google's pricing pages list video-with-audio SKUs, but the current Agent Platform documentation marks sound generation as unsupported for the standard and Fast -001 endpoints while supporting it on Lite. Verify the exact platform and callable route before budgeting for audio.
Kling 3.0 API Pricing: 720p, 1080p, and Native Audio
Kling's developer video pricing currently lists these per-second equivalents:
| Kling route | Configuration | 720p | 1080p |
|---|---|---|---|
| Kling 3.0 | No native audio | $0.084/s | $0.112/s |
| Kling 3.0 | Native audio, no voice control | $0.126/s | $0.168/s |
| Kling 3.0 Turbo | Native audio | $0.112/s | $0.14/s |
The Kling Video 3.0 model guide documents native audio, multi-shot generation, multilingual support, and outputs of up to 15 seconds.
Kling pricing changes with resolution, native audio, voice control, and Turbo versus full-model routing. Store the exact route and submitted parameters with every job.
BytePlus Seedance API Pricing: Seedance 2.0 Today, Seedance 2.5 Coming Soon
ByteDance introduced Seedance 2.5 as the next generation of the Seedance video family. Compared with Seedance 2.0, the new model expands single-generation output from up to 15 seconds to up to 30 seconds and supports larger multimodal reference sets for longer storytelling, continuity, and editing workflows. ByteDance says Seedance 2.5 can accept up to 30 images, 10 video clips, and 10 audio clips in one task.
For full capability details, see ByteDance's official Seedance 2.5 announcement.
CometAPI currently lists Seedance 2.5 API access as coming soon. Until the production route, supported parameters, and live billing are confirmed, Seedance 2.5 should not be assigned a fixed per-second price in this comparison.
Current Seedance 2.0 API Pricing
BytePlus does not publish one universal Seedance 2.0 price per second. Its ModelArk pricing page uses configuration-dependent, token-metered pricing and provides per-video ranges based on input length, model variant, and output resolution.
| Model | 480p | 720p | 1080p | 4K |
|---|---|---|---|---|
| Seedance 2.0 Mini | $0.19–$0.42 | $0.41–$0.91 | Not supported | Not supported |
| Seedance 2.0 Fast | $0.30–$0.66 | $0.64–$1.43 | Not supported | Not supported |
| Seedance 2.0 | $0.39–$0.86 | $0.84–$1.86 | $2.06–$4.57 | $4.20–$9.33 |
These figures are per-video ranges for video-input workloads, not universal per-second rates. The lower end corresponds to shorter inputs, while longer inputs and higher resolutions increase the final charge. Use the live BytePlus calculator or provider-reported usage when preparing a production budget.
For current implementation details, read CometAPI's Seedance 2.0 API Integration Guide. Teams planning to test the next generation can follow the Seedance 2.5 model page on CometAPI for availability updates.
Runway API Pricing: Gen-4 Turbo and Gen-4.5
Runway developer credits cost $0.01 each. Its current API pricing documentation lists:
| Runway model | Credits per second | USD per second | 10-second equivalent |
|---|---|---|---|
| gen4_turbo | 5 | $0.05 | $0.50 |
| gen4.5 | 12 | $0.12 | $1.20 |
Runway also routes third-party models such as Veo and Seedance. Label those as gateway routes and record the selected model and realized credit cost from the response metadata.
Is the Sora API Shutting Down, and What Should Teams Migrate To?
Yes. OpenAI's deprecation schedule says the Videos API, sora-2, sora-2-pro, and listed snapshots will be removed on September 24, 2026. OpenAI does not list a direct replacement in the deprecation table.
Existing Sora users should use August and early September as a migration window. Test at least one low-cost route, one native-audio route, and one higher-quality route—such as Veo, Kling, Seedance, or Runway—using the same prompts and reference assets.
CometAPI lets teams evaluate Sora 2 alternatives through one API. Compare available routes such as Kling Video and Seedance 2.0 by output quality, latency, retries, and review effort. For migration details, see the CometAPI Sora API Guide.
What's the Difference Between First-Party API, Gateway, and Subscription Pricing?
Do not combine these pricing surfaces without labels:
| Pricing surface | What it represents | Correct use |
|---|---|---|
| First-party API | Provider's published developer rate | Baseline for cross-provider comparison |
| API gateway | Gateway price for an exact model route and parameter set | Use for the real integration budget after checking the live catalog |
| Creator subscription | Web-app credits, plan limits, and UI features | Do not convert into API economics unless developer calls are explicitly included |
Gateway billing units, aliases, and exposed capabilities may not map one-to-one to first-party endpoints. Compare the exact route and realized charge, not only the model family name.
What Does an AI Video Actually Cost in Production?
The list price is only the generation baseline. A production budget should include retries, failed or rejected clips, reference assets, editing, review, storage, and delivery.
Expected cost per usable clip =
base generation cost × average attempts per accepted clip
- input, audio, editing, storage, and review costs
Average attempts per accepted clip = 1 ÷ acceptance rate
Example planning scenario:
| Assumption | Value | Calculation |
|---|---|---|
| Generation baseline | $1.20 per normalized 10 seconds | Example rate |
| Acceptance rate | 60% | Team assumption |
| Attempts per accepted clip | 1.67 | 1 ÷ 0.60 |
| Generation cost per usable clip | $2.00 | $1.20 × 1.67 |
| Human review | $1.00 | 2 minutes at $30/hour |
| Storage and transfer | $0.02 | Team assumption |
| Total cost per usable clip | $3.02 | $2.00 + $1.00 + $0.02 |
The same $1.20 baseline changes quickly as acceptance rate falls:
| Acceptance rate | Attempts per accepted clip | Generation cost per usable clip |
|---|---|---|
| 80% | 1.25 | $1.50 |
| 60% | 1.67 | $2.00 |
| 40% | 2.5 | $3.00 |
Track acceptance rate separately for product shots, people, dialogue, visible text, camera motion, and multi-shot scenes. A blended average can hide the prompt classes that create most retries.
Developer tip: To benchmark multiple routes without maintaining a separate client and billing workflow for each provider, review the available video models in the CometAPI model catalog. Keep first-party rates as the neutral comparison baseline, then use the gateway's live usage records for your actual integration budget.
How Do Resolution, Audio, and Inputs Change the Bill?
| Provider family | Lower-cost configuration | Higher-cost configuration | Difference |
|---|---|---|---|
| Sora 2 Pro | 720p at $0.30/s | 1080p at $0.70/s | 1080p is about 2.33× the 720p rate |
| Veo 3.1 Fast | 720p video only at $0.08/s | 720p video + audio at $0.10/s | Audio SKU adds $0.02/s where supported |
| Kling 3.0 | 720p without native audio at $0.084/s | 720p native audio at $0.126/s | Native audio is 50% higher |
| Seedance 2.0 | Lower-resolution, shorter-input range | 1080p or 4K video-input range | Token usage and total price rise materially |
Input assets matter most on token-metered or reference-heavy routes. Do not reuse a text-to-video estimate for image-to-video or video-reference jobs unless the provider confirms the same billing rule. Also record the terminal state and reported charge for failed, moderated, cancelled, or timed-out jobs.
Which AI Video API Fits Each Workload?
There is no universal “best AI video API” without testing the target workload. The following routes are starting candidates based on documented pricing and capabilities, not a quality ranking.
| Workload | First routes to test | Why | What to verify |
|---|---|---|---|
| Low-cost silent drafts | Veo 3.1 Lite; Runway Gen-4 Turbo | Lowest directly comparable entry rates | Preview access, prompt adherence, and acceptance rate |
| Fast creative iteration | Veo 3.1 Fast; Runway Gen-4 Turbo; Kling 3.0 Turbo | Speed-positioned or cost-efficient routes | Queue time, retries, and visual consistency |
| Native-audio clips | Veo audio routes where the endpoint supports them; Kling 3.0; Seedance 2.0 | Integrated audiovisual options | Lip sync, language support, endpoint availability, and audio billing |
| Reference-driven ads | Kling 3.0; Seedance 2.0; supported Veo routes | Image, video, or multimodal reference workflows | Subject fidelity, input charges, moderation, and callback reliability |
| Higher-resolution delivery | Veo 3.1; Seedance 2.0; Kling 3.0 | 1080p or 4K tiers are documented | Delivered resolution, compression, editing time, and usable cost |
| Existing Sora integration | Sora 2 plus at least two replacement candidates | Creates a controlled migration benchmark | Finish migration before September 24, 2026 |
How to Test AI Video APIs Before Production
Use a fixed evaluation set and record the actual billed response. A practical first pass is 30 jobs:
- Ten text-to-video prompts covering people, products, camera motion, visible text, and multi-subject scenes.
- Ten image-to-video jobs using the same licensed reference assets.
- Ten workload-specific jobs, such as ads with audio, product demonstrations, loops, or multi-shot sequences.
Hold duration, resolution, aspect ratio, references, audio settings, seed behavior, and reviewer rubric constant wherever the APIs allow it. Compare supported durations directly and normalize price separately rather than stitching clips to force a 10-second test.
| Field to log | Why it matters |
|---|---|
| Provider, route, model ID, and version date | Prevents comparisons across different releases or aliases |
| Submitted parameters and input assets | Reconstructs resolution, audio, and reference costs |
| Task ID and terminal state | Separates completed, failed, cancelled, and moderated jobs |
| Provider-reported usage and charge | Measures the invoice rather than estimating from submissions |
| Queue and generation time | Shows interactive versus batch suitability |
| Reviewer result: accepted, fixable, or rejected | Creates the usable-output denominator |
| Retry and fallback reason | Explains why a cheap route becomes expensive |
| Review and editing minutes | Exposes labor hidden by the model price |
| Output-copy timestamp | Prevents loss when temporary URLs expire |
Report acceptance rate, cost per accepted clip, p50 and p95 completion time, technical failure rate, moderation rate, fallback rate, and reviewer minutes for each prompt class. Run at least two passes because generation is stochastic.
Before production volume, add idempotency, webhook retries, timeout handling, per-job cost ceilings, version monitoring, and automatic copying of temporary outputs into durable storage. The guide to integrating asynchronous AI video APIs into a SaaS application covers the surrounding workflow.
Testing Multiple Video APIs With CometAPI
A multi-model gateway can reduce separate authentication, billing, and provider-client work. Use the CometAPI video model catalog to identify available routes and the live CometAPI pricing page for gateway-specific budgeting.
Before choosing a route, verify the live model ID, inputs, duration, resolution, audio mode, callback behavior, failure billing, output retention, and lifecycle status. Use first-party rates as the neutral comparison baseline and actual gateway usage records for the integration budget.
FAQ
How much does an AI video API cost in 2026?
Directly comparable first-party rates in this guide range from $0.03 to $0.70 per generated second. Production cost increases when retries, review, editing, storage, and rejected clips are included.
Which AI video API is cheapest?
Veo 3.1 Lite at 720p without audio has the lowest fixed first-party rate in the main comparison at $0.03 per second. It is the cheapest production choice only if its availability, supported duration, and accepted-output rate meet the workload.
How much does a 10-second AI video cost?
Normalized examples range from $0.30 for Veo 3.1 Lite at 720p without audio to $7.00 for Sora 2 Pro at 1080p. Because some APIs do not support an exact 10-second request, calculate the invoice from an allowed duration.
Is the Sora API being shut down?
Yes. OpenAI will remove the Videos API, sora-2, and sora-2-pro on September 24, 2026. Its deprecation table does not list a direct replacement.
What should developers test before selecting a video API?
Test identical prompts, inputs, durations, resolutions, aspect ratios, and audio settings. Measure actual charge, accepted-output rate, retries, failures, moderation, p50/p95 completion time, reviewer effort, output retention, and model lifecycle.