AI video pricing index: model rates and cost examples (2026)

For a Hailuo-specific purchase, see Hailuo 2.3 pricing. For spoken audio, the ElevenLabs pricing guide explains credits and voice-cloning access separately from video-generation charges.

Magic Hour publishes this price index and provides access to some of the listed models. We record the exact model, duration, resolution, audio setting, and provider shown in each linked primary source; rates are not normalized into quality or value rankings. Prices were checked September 9, 2026.

Verified direct API examples

USD prices exclude account-specific discounts, taxes, and services outside the quoted generation. These rows are not equivalent quality or feature claims. A provider may support different durations or input modes; consult its model documentation before assuming a particular combination is available.

For detailed plan and credit breakdowns, review Pika pricing, Runway pricing, Runway Gen-4.5 pricing, Seedance 2.0 pricing, HeyGen pricing, VEED pricing, and the Sora migration guide. Each page separates subscription invoices, credits, API billing and accepted-output costs.

For the other major consumer platforms, see Kling pricing, Luma pricing, PixVerse pricing and Invideo pricing. Compare the actual invoice and settings rather than treating different providers’ credits as equivalent. The Vidu AI review includes current Q3 direct-API rates and a cost-per-accepted-clip example.

A defined job: fifty accepted clips

Suppose the deliverable is fifty accepted six-second 768p clips and you budget three attempts per accepted clip. That means 150 generations. The published Hailuo 2.3 Fast rate gives a hypothetical $28.50 generation bill; Hailuo 2.3 gives $42.

The three-attempt allowance is an assumption, not a measured success rate. At the listed $0.19 Hailuo 2.3 Fast generation rate, the same fifty-clip job would cost $9.50 at one attempt per accepted clip, $19 at two, or $28.50 at three. These scenarios isolate generation charges; they do not establish equivalent quality or a future acceptance rate.

A worksheet for your own video budget

If no output meets the brief, report zero accepted outputs rather than a cost of $0 per usable clip. The commercial image-to-video dataset records charged credits and technical completion; acceptance costs remain unknown until the outputs are scored against a buyer's requirements.

How should you price a multi-model platform?

Use the platform’s quoted cost for the selected model and settings. Its credit unit, discounts, and included workflows may differ from the model developer’s direct API. Magic Hour’s API documentation and pricing page are the starting references for its own service.

MiniMax API pricing page, captured September 25, 2026. Billing selection, currency and any promotion are shown in the screenshot; recheck the live page and checkout before buying. View official source

Google Gemini API pricing page, captured September 25, 2026. Billing selection, currency and any promotion are shown in the screenshot; recheck the live page and checkout before buying. View official source

For subscription purchasing, see the AI video generator pricing guide. Keep the full invoice and annual commitment visible when allocating subscription spend to outputs.

What costs are easy to miss?

• Additional attempts that do not produce an accepted clip.
• Audio, higher resolution, or separate editing and enhancement operations.
• Application storage and delivery outside generation charges.
• Unused subscription capacity and credits that expire.
• Team seats, taxes, and renewal-price changes.

Why is Sora absent from the active comparison?

OpenAI lists September 24, 2026 as the Sora API discontinuation date, after ending the consumer app and web experience in April. See OpenAI’s notice. Remaining historical rates do not justify recommending a new long-term integration.

How should this index be cited?

Include the verification date, provider, model, settings, and billing unit. Link the original pricing source with any quoted figure. Do not cite this table as a quality benchmark or as proof of a provider’s total production cost.

Continue learning

Continue learning: use the complete AI video workflow to test shortlisted models, then diagnose weak outputs with the prompt troubleshooting guide.