OpenAI killed Sora. Not paused, not deprioritized, killed. The product that put AI video on every timeline in 2024 went dark for consumers in late April, and the API is on a clock that runs out in September.
Let that sink in for a second.
The model everyone benchmarked against, the one that made "is this real?" a normal question to ask about a clip of a fake dog, got shut down by the company that built it. Reporting tied the decision to compute shortages and cost pressure, with some outlets claiming the service was reportedly bleeding around a million dollars a day. OpenAI didn't give a clean reason in its shutdown notice. The numbers floating around are reported, not confirmed, so take them as the rough shape of the thing rather than gospel.
Here's the thing though. Sora leaving didn't slow the space down at all. If anything it cleared the runway for everybody else.
The Big Story: Audio Stopped Being Optional
A year ago you generated a clip and then went hunting for sound. Music here, a sound effect there, lip-sync glued on in a separate tool. It was a pipeline, and it was annoying.
That's mostly over now. Most of the leading models generate synchronized audio in the same pass as the picture. Dialogue that matches the mouth. Footsteps that land when the foot lands. Ambient room tone that fits the room. We went from basically nobody doing this to most of the field doing it, and it happened fast.
If you only remember one thing from this month's roundup, remember that one. It changes the whole workflow.
Who's On Top Right Now
Two names keep trading the lead depending on what you're trying to do.
Google Veo 3.1
Veo is the all-rounder. It's the one people reach for when they want a clip to just look correct on the first try. Strong prompt adherence, native audio, and 4K output in both landscape and portrait. Google shipped a cheaper Lite tier this spring and trimmed prices on the Fast tier, which tells you they're feeling the pressure from the cheaper challengers and don't intend to lose on cost alone.
You can get at it through Google's consumer plans or through the API. The consumer route runs about twenty bucks a month on the mid tier, with a much pricier top plan if you want the good defaults and more headroom. On the API side you're paying somewhere in the range of a few cents per second up to roughly forty cents a second once you turn on audio, with 4K sitting at a premium above that. Numbers move, so treat those as signals not contracts.
Kling 3.0
Kuaishou's Kling jumped to version 3.0 in early February and it's the value story of the season. The headline feature is multi-shot storyboarding, where you define a sequence of camera cuts (up to six) in a single generation instead of stitching separate clips together by hand.
Pair that with native multilingual audio, lip-synced dialogue across several languages, a motion brush, and 4K output, and you've got a tool that does in one pass what used to take an afternoon. Clips run up to around fifteen seconds. Pricing lands in the neighborhood of a dime per second depending on mode.
My honest take? If you're a solo creator and you want the most finished-looking output per dollar this month, Kling 3.0 is the one to try first. Veo is the safer pick for clean narrative shots, but Kling's storyboard tool is the most genuinely new thing anyone shipped this season.
The Rest Of The Field
The crown matters, but the supporting cast is where a lot of the interesting work happens.
ByteDance's Seedance 2.0 went global in the spring after a China-first launch. Its trick is reference stacking, you can feed it a pile of images plus video and audio clips and it'll hold consistency across all of them while generating picture and sound together. It pushes past 1080p natively and handles clips up to roughly fifteen seconds.
MiniMax's Hailuo is the budget champion. It defaults to 720p and reaches 1080p on paid tiers, clips are shorter (think under ten seconds), and the per-second cost undercuts almost everyone. If you're iterating on a hundred variations to find one good one, cheap-and-fast beats slow-and-perfect every time.
Runway Gen-4.5 is still the pro's control surface. Motion brush, scene consistency, reference-driven characters, and a real toolkit around text-to-video, image-to-video, video-to-video, and keyframes. When you need to direct the camera instead of just describe the scene, this is where directors land.
Luma's Ray3 leans into image quality with native HDR, which is rare, plus a Modify Video feature that reworks existing footage instead of starting from a blank prompt. On the open side, Wan keeps shipping (first and last frame control, multi-image input, very long prompts), and Shengshu's Vidu stays strong for animated and series-style work with longer audio-video runs.
Not every model fits every job. That's kind of the point now.
The Lineup At A Glance
Here's the live field, roughly where it sits in early June. Sora's missing on purpose, the consumer product is gone and the API is winding down, so it's not something to build on anymore.
| Model | Who makes it | Resolution and length | Rough price signal | Best for |
|---|---|---|---|---|
| Veo 3.1 | Up to 4K, short clips | A few cents to ~40c per second with audio | Clean narrative shots on the first try | |
| Kling 3.0 | Kuaishou | 4K, up to ~15s | Around a dime per second | Multi-shot storyboards with native audio |
| Seedance 2.0 | ByteDance | Past 1080p, up to ~15s | Low cents per second | Reference stacking and consistency |
| Hailuo | MiniMax | 720p, 1080p on paid, under ~10s | The cheapest of the bunch | Fast, high-volume iteration |
| Gen-4.5 | Runway | Short clips, deep editing tools | Subscription plus credits | Directing camera and motion |
| Ray3 | Luma | Up to 4K HDR, ~10s extendable | Subscription plus API | HDR color and editing existing footage |
If you want to go deeper on the three that matter most for most people, we broke down our Sora vs Kling vs Runway comparison in its own piece, and it holds up even with Sora on its way out, because the framework for judging these tools didn't change.
What This Means If You're Actually Making Stuff
Strip away the leaderboard drama and the practical picture is simple.
Pick by job, not by hype. Here's the cheat sheet:
- A polished hero shot for a landing page? Veo 3.1 or Kling 3.0.
- A multi-shot scene without touching an editing timeline? Kling's storyboard.
- Fifty cheap drafts before you commit to one? Hailuo.
- Directing camera and motion like an actual director? Runway Gen-4.5.
- HDR color that pops on a good display? Luma Ray3.
- Owning the whole pipeline on your own hardware? An open model like Wan.
And budget for waste. None of these nail it first try every time. You'll reroll. You'll trim. The real cost of a finished clip is the cost-per-second times the number of attempts it takes to get one you'd actually publish, and that multiplier is usually three or four, not one.
The tooling has finally caught up to the ambition. A year ago "make a 15-second scene with synced dialogue and a couple of camera angles" meant four tools and a lot of patience. Now it's one prompt and a coffee.
The most underrated shift this year isn't resolution. It's that the boring glue work, the audio sync, the shot stitching, the lip matching, got absorbed into the model itself. That's what actually saves you time.
If you're shopping the broader landscape and not just video, the best AI tools roundup covers the writing, design, and automation pieces that usually sit on either side of a video workflow.
The Open Models Are Quietly Catching Up
One thread that doesn't make the headlines but matters a lot if you care about owning your pipeline: the open-weight side keeps closing the gap.
Wan and a handful of other open models now ship features that used to be exclusive to the paid frontier, like first and last frame control and multi-image conditioning. They're not beating Veo or Kling on raw polish. They don't have to. If you can run a model on your own hardware, the per-clip cost drops toward the cost of electricity, and nobody can pull the product out from under you the way OpenAI just did with Sora.
That last point isn't theoretical anymore. The Sora shutdown taught a lot of teams a real lesson about building a business on top of a model they don't control. Open weights are insurance against exactly that.
The Trade You're Making
Open isn't free, to be clear. You're trading a clean web app for a setup that wants a real GPU and some patience. You'll fiddle with weights and configs. You'll wait longer on slower hardware.
For most casual creators that trade isn't worth it, and a hosted model is the right call. But for studios and high-volume shops, the math flips quickly once you're rendering thousands of clips a month.
A Few Honest Caveats
Every roundup like this comes with fine print, so here's ours.
The benchmark leaderboards are noisy. Different sites rank these models in different orders depending on what they tested and who voted, so treat any single ranking as one opinion rather than the truth. The prices shift constantly, often downward, sometimes overnight, which is great for you and annoying for anyone trying to write them down. And "best" depends entirely on the shot you're making, a tool that wins on a product demo can lose badly on a fast-action scene.
So don't over-index on a chart. Run two or three of these on your actual use case and trust your own eyes over anyone's arena score, including ours.
Looking Ahead
The obvious open question is whether anyone fills the gap Sora left at the top of public attention, or whether the field just stays this crowded and competitive with no single king. Cheaper tiers keep landing, audio keeps getting better, and the length ceiling keeps creeping up. None of that looks like slowing.
This space moves weekly, so treat every number here as a snapshot of early June, not a permanent fact. We'll refresh this roundup every month and track what actually shipped versus what got announced. See you in July.