What Actually Happened
The frontier video models stopped being scarce. A single Runway subscription on its Standard plan, somewhere around $12 to $15 a month, now bundles access to Gen-4.5 plus integrated third-party models including Google Veo 3.1 and Kling 3.0 Pro, along with Seedance and FLUX. That is multiple frontier models under one cheap subscription, sitting behind one login (Somake).
Meanwhile the high end keeps climbing. Kling's top "Ultra" tier rose from $128 a month at its August 2025 launch to $180 a month by January 2026, a 41% increase in roughly six months, and it is now monthly-only (eesel).
And the per-second economics are racing to the floor underneath all of it. Google Veo 3.1 Lite is priced at about $0.05 per second, undercutting Kling's roughly $0.07 (MindStudio).
Read those three facts together and a pattern shows up. The same models are available to everyone, the prices at the top are being defended through bundling and hikes rather than exclusivity, and the cost of a generated second is now measured in pennies. None of that describes a moat. It describes a utility.
None of this means the models are bad. Veo 3.1 renders genuinely impressive footage, and Kling's longer-duration clips solve real problems. The point is narrower and harder to dodge: a capability that everyone can buy this afternoon for the price of a team lunch cannot be the thing that sets your video apart. Scarcity is what makes a tool a differentiator, and the scarcity is gone.
So Stop Vendor-Shopping by Model Name
If a tool costs $15 a month and your competitor down the street can subscribe to the exact same thing this afternoon, then "we use the best AI video model" is not a position. It is a receipt. That is the contrarian part most studios will not say out loud.
We hear the model question on most intro calls. A marketing leader asks whether we run Veo or Runway or Kling, the way you might ask a contractor which brand of drill they own. It is a reasonable instinct and a useless filter, because the honest answer is that we run whichever one produces the shot, and so does everyone else worth hiring.
A lot of buyers are still evaluating production partners the way they'd evaluate a software license, by asking which model is under the hood. That made sense in a world where access was gated and expensive. It makes no sense in a world where access is a commodity rented by the minute.
When a capability commoditizes, the value doesn't disappear. It moves. It moves to the layer above the tool, the judgment about what to point the tool at, and the layer after the tool, the system that gets the output in front of the right people enough times to matter.
Those two layers, taste and distribution, are exactly the things you cannot add to a cart. You can't subscribe to a point of view. You can't expense an audience. That's the whole game now.
The Data From Our Retainer Book
The numbers in our own shop make the case plainly, and we did not back into this thesis to fit it.
Across our retainer book, the share of a project budget that goes to which generative tool we use sits in the low single digits. The overwhelming majority goes to direction, on-camera talent, and the edit-and-distribution system. The tool line item is a rounding error next to the work that actually decides whether the video lands.
Put concrete numbers on it and the gap is almost comic. On a typical B2B explainer, the generative tool might account for a few dollars of render and subscription cost. The direction, the on-camera talent, the scripting, and the cutting that follow run into the thousands. Spending your evaluation energy on the few-dollar line while ignoring the few-thousand-dollar one is how teams end up with technically flawless video that nobody watches.
We tested the other direction too. In a recent internal production time-study, swapping the generative model on a B2B explainer changed our render cost by a few dollars and moved zero of the variables that actually drove completion rate or pipeline. The expensive part of the video was never the model. It was the brief, the talent, and the cadence behind it.
The market keeps proving the same point at a larger scale. OpenAI discontinued the Sora app and web on April 26, 2026, with the API set to sunset on September 24, 2026, after reportedly burning roughly $15M a day against about $2.1M in lifetime revenue (eWeek). If a category-defining model can't hold a moat, the model you rent for $15 a month certainly isn't one for you.
The Honest Counter-Argument
The strongest case against all this is simple, and it's partly right. Better models do produce better output. A studio on the current frontier can hit a fidelity, a coherence, and a prompt-adherence that a studio on last year's model can't, and that gap shows up on screen. Access also matters for speed and cost. The shop running the cheaper per-second model ships more iterations per dollar, and iteration is real leverage.
All true. Here's why it doesn't rescue the moat. The frontier is shared. The same bundle that gives you Veo and Kling and Gen-4.5 gives your competitor the same three. Quality gaps between frontier models are real for a few months and then they close, because everyone re-subscribes the day the new version drops.
So model quality is table stakes, not advantage. You should absolutely run good models, the way a restaurant should absolutely run a working oven. Nobody picks the restaurant for the oven. The speed-and-cost edge is real but small, a few dollars and a few minutes per asset, and it's available to anyone willing to read the pricing page. What's not available to anyone is the taste to know which iteration is the right one and the distribution to make sure it's seen.
What to Do Monday
First, change how you evaluate partners. Stop asking a production studio which model they use and start asking how they decide what to make, who's directing it, and what happens to the video after it renders. The answer to those three questions is the actual product. The model is just plumbing.
Second, keep your tools interchangeable on purpose. Don't architect your process around one vendor's interface or one model's quirks. When access is a commodity and prices swing 41% in six months, the studio that can swap the model without changing the work is the studio that stays cheap and current. Build for substitution.
Third, move the budget to where the value actually is. If the tool is low single digits of your spend, that's correct, leave it there. Put the real money into direction, talent, and the edit-and-distribution system, because that's the part the data says decides completion and pipeline.
Fourth, build a distribution system, not a render queue. Generating the video is the easy, cheap, commoditized part now. Getting it in front of the right buyers at the right cadence, across the right channels, enough times to move a deal, is the part nobody can rent. That's where a B2B video program either earns its keep or doesn't.
Fifth, audit your own content the way a stranger would. The uncomfortable test is to mute the vendor names and ask whether anyone outside your company can tell which tool made the video. If they cannot, and they almost never can, then the tool was never the story. The story was whether the video said something worth hearing and reached someone who needed to hear it.
Do those four things and the price of the model becomes exactly what it should be, a footnote.