AI video model benchmark
One brief, twelve engines from free to flagship. Every clip published, including the wobbly ones, with what it cost and how long it took.
The run
- Models attempted
- 12
- Delivered a file
- 11
- Failed outright
- 1
- Whole run cost
- 1540 cr ≈ $4.31
- Fastest
- Luma Ray Flash 2 · 41.9s
- Run on
- 12 September 2026
Model comparisons on the internet are mostly cherry-picked. Someone runs a prompt twenty times, keeps the best frame, and publishes it next to a competitor's first attempt. That tells you nothing about what happens when you type your own brief.
So this is the opposite: one prompt, written once, sent to twelve text-to-video engines spanning every price tier in the studio — free pool, fast, premium and flagship — one attempt each, one at a time. The clip below each model is the first and only thing it returned. The credit figure is this studio's list price, the time is measured from submit to a finished file, and the model that failed is still in the table.
The brief, verbatim
A barista pours steamed milk into a flat white on a marble counter. Morning light through a window behind, steam rising from the cup, slow push-in, shallow depth of field, photographic.
Settings: 16:9, five seconds where the engine allows it, 1080p where it is offered. One attempt per model, run one after another, no re-rolls and no per-model prompt tuning.
The numbers
What each engine charged and how long it took.
Wall-clock from submit to a finished file on one account, single run. Credits are the studio's list price, which is what the account was charged.
| Model | Price | Time | Delivered | Result |
|---|---|---|---|---|
| Veo 3.1 Fast (free pool)withdrawn | Free | 239s | 4s clip · 3.9 MB | Delivered |
| Kling 3.0 (free pool)withdrawn | Free | 267.9s | 5s clip · 9.2 MB | Delivered |
| Seedance 2 Fastwithdrawn | 60 cr | 94.8s | 5s clip · 2 MB | Delivered |
| Veo 3.1 Fastwithdrawn | 60 cr | 74.1s | 4s clip · 2.8 MB | Delivered |
| Hailuo 2.3 Fastwithdrawn | 60 cr | — | — | Provider failure |
| Luma Ray Flash 2withdrawn | 60 cr | 41.9s | 5s clip · 0.7 MB | Delivered |
| Kling 3.0withdrawn | 140 cr | 339.9s | 5s clip · 7 MB | Delivered |
| Runway Gen4.5withdrawn | 140 cr | 95.1s | 5s clip · 1.9 MB | Delivered |
| Wan 3.0withdrawn | 140 cr | 159.3s | 5s clip · 13.7 MB | Delivered |
| Seedance 2.5withdrawn | 140 cr | 166s | 5s clip · 4.7 MB | Delivered |
| Veo 3.1withdrawn | 400 cr | 81.1s | 4s clip · 2.4 MB | Delivered |
| Sora 2 Prowithdrawn | 400 cr | 127.2s | 4s clip · 1.7 MB | Delivered |
One of the 12 engines took the job and then returned a failure from the provider. Credits are refunded automatically when a render fails, so those rows cost nothing — they stay in the table because that is the part of a benchmark nobody else publishes. 12 of them have since been withdrawn from the studio altogether: the failure was the provider's, it kept happening, and no backup renderer carries them — so they are no longer offered rather than left in the picker to fail again.
Every output
The actual files, unretouched.
First and only attempt from each engine. Open any one to see its full run page.
Free · 239s · 3.9 MB
Free · 267.9s · 9.2 MB
60 cr · 94.8s · 2 MB
60 cr · 74.1s · 2.8 MB
60 cr · 41.9s · 0.7 MB
140 cr · 339.9s · 7 MB
140 cr · 95.1s · 1.9 MB
140 cr · 159.3s · 13.7 MB
140 cr · 166s · 4.7 MB
400 cr · 81.1s · 2.4 MB
400 cr · 127.2s · 1.7 MB
What it showed
Reading the run.
- Nothing in the slate misread the brief. Every delivered clip is a barista pouring steamed milk into a cup in morning light, and in almost all of them the latte art forms by the final frame — these engines have seen a great deal of coffee.
- Price did not buy speed. The 400-credit Veo 3.1 delivered in 81 seconds; the 140-credit Kling 3.0 took 340, the longest wait in the run; and the cheapest paid engine, Luma Ray Flash 2 at 60 credits, was done in 42.
- The free pool is not a lesser model, it is a longer queue. Free Veo 3.1 Fast returned a comparable four-second clip in 239 seconds where the 60-credit route took 74 — and free Kling 3.0 (268s) actually beat the 140-credit Kling 3.0 (340s) on this run.
- File weight varies twentyfold for the same five seconds: Wan 3.0 handed back 13.7 MB, Luma Ray Flash 2 just 0.7 MB. If you are grading or cutting the clip afterwards, that is a real difference in what you have to work with.
- One engine failed outright: Hailuo 2.3 Fast took the job and returned no clip after 68 seconds. It has since been withdrawn from the studio — the failure repeated and no backup renderer carries it — so that row has no model page to open. Hailuo 2.3 itself is unaffected and still in the picker.
- Four and five seconds is the unit. Veo and Sora take four, six or eight; the rest take five or ten. Nothing here makes a thirty-second spot in one render, so a longer piece is several of these cut together.
FAQ
What people ask.
In this run, between 42 seconds and 5 minutes 40 for a single clip of four or five seconds, with the middle of the field landing just over two minutes. The wait is provider queue plus render, and it does not track price: the 400-credit engine came back in 81 seconds while a 140-credit one took 340. Plan on a couple of minutes a clip and draft on the fast engines rather than waiting on the flagship.
Every engine that delivered understood the brief — a barista pouring milk, morning light, slow push-in — so "best" comes down to taste and to what you are cutting it into. The clips are embedded above at full quality precisely so you can judge that yourself: look at the jug hand, the milk stream and how the latte art forms, because that is where these engines separate. What the table settles is the part nobody publishes: what each one charges and how long it makes you wait.
On this studio, nothing at all on the free pool, 60 credits for a fast clip, 140 for a premium one and 400 for a flagship — about 17 cents, 39 cents and $1.12 at Starter pricing. Eleven delivered clips across every tier came to 1,540 credits, roughly $4.31 for the whole run.
Yes — two of the clips above cost zero credits. A handful of engines run on a free experimental pool with a longer queue and a daily allowance, and on this run the free route returned a comparable clip to the paid one from the same family. Free accounts can also spend their daily credits on the 60-credit fast engines; the 140- and 400-credit flagships need a paid plan.
No. One attempt per engine, one after another, no re-rolls and no per-model prompt tuning. Every clip that came back is published above and on the public wall, including the wobbly ones, and the engine that returned nothing is still in the table.
Method
How this was run, so you can argue with it.
Each model received the identical prompt above through the same studio a customer uses, one model at a time so nothing queued behind anything else. Time is wall-clock from the moment the request was accepted to the moment a playable file existed — provider queue included, because that is the wait you actually experience. It is a single run per model, not an average, and a second run would land somewhere slightly different.
Every delivered file is shown unedited and keeps its own run page, which is what the thumbnails above link to. Nothing was upscaled, re-rolled, colour-corrected or re-prompted per engine. The credit figures are this studio's list prices, and the total is what the delivered renders cost — a render that fails refunds itself, so the failures above are free in money and expensive only in time.
Run your own brief instead
Every model still offered in this table is in the same picker, on one balance. Free account, 22 credits a day, no card.