Veo 3 pricing in 2026: every tier, API rate, and what AI video actually costs

On April 26, 2026, OpenAI shut down the Sora app. On September 24, 2026, the Sora API followed it, per OpenAI’s own discontinuation notice. Teams that built video features on a $0.10 per second model got six months’ notice to migrate. OpenAI has named no successor, which puts Google Veo at the top of nearly every migration shortlist.

That migration is why Veo pricing suddenly matters to finance teams and not just creators. AI video went from experiment to line item, and the vendor that owned the category just proved it can vanish.

If your company generates video with AI, you now have two jobs: know exactly what Veo costs, and know what your usage is worth. Let’s do both.

How much does Veo 3 cost in 2026?

As of September 2026, Veo costs $0.05 to $0.60 per second of finished video through the Gemini API, or comes bundled into Google’s consumer AI plans, which run from the entry-level Google AI Plus tier up to Google AI Ultra at $199.99 per month.

The current model family is Veo 3.1, generally available since November 17, 2025 after an October 2025 preview, with the Lite tier added in March 2026. It comes in three API tiers: Lite, Fast, and Standard. Veo 3.1 is the current release in the Veo 3 family, so Veo 3 pricing in practice means Veo 3.1 rates, and every figure below is a Veo 3.1 rate. Audio is included in every rate, and Google only bills successful generations.

Here is the full Gemini API rate card, as of September 2026:

List prices from each provider’s first-party API rate card, checked September 30, 2026.

Three details in that table decide real budgets. First, Standard costs the same at 720p and 1080p, so there is no savings in dropping resolution on your hero shots. Second, Lite has no 4K path at all. Third, the jump from Fast to Standard is 4x at 720p, which makes tier selection the single biggest cost lever you control.

Every clip maxes out at 8 seconds per generation. Longer videos mean chaining generations, and the meter runs on each one. A 32-second product video is four Standard generations, or $12.80 at 1080p, before a single retake.

Google’s consumer plans bundle Veo access into monthly Flow credits, the prepaid units those plans spend inside Flow, Google’s consumer video app, rather than per-second billing. Google’s plans page now runs three tiers: entry-level Google AI Plus, Google AI Pro at $19.99 per month, and Google AI Ultra at $99.99 or $199.99 per month, after Google cut its top tier from $249.99 at Google I/O 2026. Pro includes 1,000 monthly Flow credits; the two Ultra tiers include 10,000 and 25,000.

Credits translate to videos at different burn rates by tier. On Pro, 1,000 credits buys approximately 100 Lite videos, 50 Fast videos, or 10 top-quality videos, Flow’s highest-fidelity setting, per month. Top-tier Ultra’s 25,000 credits stretch to roughly 250 top-quality generations; the $99.99 tier’s 10,000 stretch to about 100. College students currently get a year of Google AI Plus at no cost, which explains a good share of the Veo content flooding your feeds. Availability varies by country.

The subscription math has a crossover point worth calculating. Spend Pro’s entire allowance on video and 10 quality clips cost $19.99, about $2 each, though those credits also feed every other AI feature on the plan. The same 8-second clip on the Standard API costs $3.20. Light, unpredictable usage favors the subscription. Anything programmatic, automated, or embedded in a product belongs on the Gemini API, where you pay per second and can meter what’s happening.

Worth knowing before you commit either way: Veo access through Claude, ChatGPT, and other AI subscriptions does not exist. Google keeps Veo exclusive to its own plans and APIs, so there is no side door through a tool you already pay for. Budget for it as its own line.

That’s the pattern across all of AI: the subscription is the visible tip, and API usage is the iceberg underneath. It’s the same dynamic we’ve documented with ChatGPT’s plans, where seats are predictable and the API becomes the real bill.

What does the Veo 3 API cost per video?

A single Veo API video costs between $0.40 and $4.80 depending on tier, resolution, and length, and the arithmetic is simple. The Veo 3.1 API bills per second of output, and rates differ between the Gemini API and Vertex AI, so the card above is the Gemini API surface.

Worked examples at the maximum 8-second length:

Getting access requires a paid Gemini API account. There is no video generation on the free tier: Gemini’s free tier covers text models, while every Veo second is billed. The one mercy in the fine print: Google states you’re only charged when a video successfully generates, so filtered or failed jobs don’t hit the invoice.

One more model deserves a line item in your planning. Google’s Gemini Omni Flash, its next-generation video model, prices out to an effective $0.10 per second at 720p on token-based billing.

Google is clearly experimenting with how video should be priced, which means today’s rate card is a snapshot. Google’s own pricing table already dates its next move: Gemini 3.8 Flash output runs $3.75 per million tokens through December 31, 2026 and $7.50 from January 1, 2027.

Why do Veo videos cost more than the rate card says?

Real Veo costs typically run 3x to 10x the advertised per-second rate, because you pay for every take and keep only one. The rate card prices a generation. Your budget pays for a workflow: prompt attempts, style iterations, extended cuts, upscales, and the clips your brand team rejects on sight.

Consider an honest accounting of one 30-second ad. Four chained Standard 1080p generations cost $12.80. Now add reality: 10 takes to nail the opening shot, five for the product close-up, and a second full version for the A/B test. A “$12.80 video” lands near $90 without anyone doing anything wrong. Multiply by a content calendar and the gap between list price and actual spend becomes the whole story.

This iteration factor isn’t unique to video, but video makes it expensive fast. Our analysis of production AI workloads found the true cost of a resolved AI task often runs 10 to 50x higher than the posted per-call price once the full stack is counted, though that figure measures resolved text-model tasks rather than video generations. With text models that stack means retrieval and orchestration. With Veo, it’s mostly your own creative standards.

The 8-second cap compounds this. A 9-second concept costs two full generations, not 1.125 of one. Teams that storyboard to 8-second beats before generating consistently spend less than teams that generate first and edit later.

How does Veo 3 pricing compare with Sora and other video models?

Veo 3.1 Standard costs 4x more per second than Sora 2 did, but Sora’s price advantage ended with the API on September 24, 2026. Comparing the models people shortlist:

The comparison every migrating team is running: Sora 2 at $0.10 per second maps almost exactly to Veo 3.1 Fast at $0.10 to $0.12. Teams that budgeted for Sora Standard can move to Veo Fast nearly cost-neutral. Teams that used Sora 2 Pro at $0.30 to $0.70 will find Veo Standard at $0.40 to $0.60 familiar, occasionally cheaper.

There’s also a budget flank worth knowing about. xAI’s Grok Imagine positions itself as the fast, low-cost video option, consistent with the aggressive pricing we covered in our Grok pricing breakdown. And OpenAI has not announced a Sora successor, which leaves Veo as the default destination for most migrating teams and removes a source of pricing pressure that existed a year ago.

The deeper lesson from the Sora shutdown belongs in every AI budget review. A model whose early demo led filmmaker Tyler Perry to pause an $800 million studio expansion lasted just under two years as a public product. Per-second price is one input; platform durability is another, and the second one just cost a lot of teams an unplanned migration. Vendor risk is now a line item.

What does Veo spend look like at scale?

At production scale, Veo becomes a five-figure monthly line item on modest volume, because a second of generated video is among the most expensive units any major AI API bills for.

One 8-second Standard clip costs $3.20, roughly what a million output tokens cost on Gemini 3.8 Flash at its current $3.75 rate. Your video feature spends like a text feature with the decimal point moved.

Run the numbers on a modest product feature: 5,000 clips per month, 8 seconds each. On Fast 1080p that’s $4,800 monthly. On Standard 1080p it’s $16,000. Apply an assumed 3x iteration factor, the low end of the range above and a planning figure rather than a vendor one, and Standard lands at $48,000 per month, or $576,000 annualized, for one feature.

Gartner forecasts worldwide AI spending will reach $2.59 trillion in 2026, up 47% year over year, and line items like this are how.

Here’s the question that separates a healthy AI budget from a broken one: is that $48,000 good or bad? By itself, the number can’t tell you. If those clips serve 2,000 paying customers, video costs $24 per customer per month. If your average customer pays $500, that’s a great trade. If they pay $30, your video feature is eating your margin, and nobody decided that on purpose.

The unit view changes the conversation entirely. The same $48,000 reads three different ways depending on what it buys:

Illustrative example. Figures assume 5,000 eight-second Standard 1080p clips per month and an assumed 3x iteration factor, which is a planning figure rather than a Google-published one. Your ratio will differ.

Total spend answered none of those questions. Unit cost answered all three. That framing shift, from “what did AI cost” to “what did each unit of value cost,” is the difference between a budget review and a budget panic.

This is where AI spend stops being an engineering metric and becomes a finance question. One CloudZero customer, an AI-powered search company, connected its billing data and discovered OpenAI alone was 25% of its total cloud spend. Not a line item anyone approved. A quarter of the bill. It accumulated, one API call at a time. Video accumulates faster.

The mechanics matter here.

Veo charges arrive through your Google billing, mixed into the same invoices as compute, storage, and everything else on Google Cloud. CloudZero ingests that billing data alongside AWS, Azure, Kubernetes, Snowflake, and direct AI platform connections to OpenAI, Anthropic, and Cursor, then allocates all of it to the dimensions your CFO asks about: team, product, feature, customer. Cost per video generated. Cost per customer served. Margin per AI feature shipped.

As Scott Castle, CloudZero’s Chief Product Officer, puts it: “Seeing what you spent is the easy part now.” The hard part, and the part that decides AI ROI, is connecting spend to what it produced. Skyscanner, Malwarebytes, and Upstart use CloudZero to do that across their cloud and AI portfolios.

If you’re standing up a Veo integration this quarter, that’s the right moment to wire in visibility. Book a demo and we’ll show you your video generation costs allocated by feature and customer, or take the self-guided tour if you’d rather poke around first.

How do you keep Veo 3 costs under control?

The teams that control Veo spend do five things consistently, and none of them require generating less video. They ladder tiers, cap resolution by use case, batch what can wait, storyboard to the 8-second grid, and measure cost per unit of business value instead of cost per month.

• Ladder your tiers. Draft on Lite at $0.05 per second, refine on Fast, and reserve Standard for approved finals. 10 generations run as nine Lite drafts plus one Standard final cost $6.80 per shipped 8-second clip. The same 10 generations on Standard cost $32. Same output, nearly 5x apart.
• Match resolution to destination. A 4K generation costs 50% more than 1080p on Standard, and most social platforms compress it back down anyway. Paying $4.80 for pixels Instagram will destroy is charity to no one.
• Batch the non-urgent. Google’s Batch API halves rates across much of the Gemini lineup in exchange for asynchronous turnaround, and third-party rate sheets apply the same 50% discount to video. Veo’s official rate card doesn’t publish batch pricing, though, so confirm the discount reaches your access path before building an overnight render pipeline around it. Half-price seconds are the closest thing to a coupon in this market, if your account has them.
• Fix prompts before scale. Every retake is a full-price generation. Teams that invest in prompt templates and reference-image workflows cut iteration counts dramatically, the same discipline that pays off in OpenAI cost optimization and every other usage-billed service.
• Measure unit cost, not total cost. Total spend rising can be great news if cost per customer is falling. CloudZero’s AI Signals, launched in September 2026, surfaces gateway AI usage, meaning the calls flowing through your AI gateway, within seconds of those calls firing, and anomaly detection across your connected billing catches the rest, so a runaway generation loop shows up in an alert instead of on next month’s invoice. Pair that with inference cost fundamentals and the broader toolkit in our guide to cloud cost management platforms, and Veo becomes a managed investment instead of a surprise.

The Sora era taught everyone that AI video capability can appear, reprice, and disappear inside 18 months. You can’t control Google’s rate card, the model roadmap, or the next deprecation notice. You can control whether you see your AI video generation costs clearly enough to make every one of those events a planned adjustment rather than a fire drill.

Get a free cost assessment and find out what your AI spend is buying.

Frequently asked questions about Veo 3 pricing

Author Spotlight

Lyne Carolyne

Lyne Carolyne has several years of experience in AI and cloud economics and brings that understanding into the content she creates. Outside work, she's an avid explorer.