Inference
Three things work together behind every price: direct partnerships with the labs, our own inference engine for the models we serve ourselves, and a router that sends each request to the fastest, most cost-effective healthy route.
Pillar 03 · Automatic routing
Our router sits in front of every model. It knows which routes are healthy right now, scores them on speed and cost for your exact request, and fails over on its own when something degrades.
Pillar 02 · Our own inference engine
For the models we host, we run our own serving stack on GPUs we schedule ourselves. Less idle time and less wasted work per output means a lower cost per second — and a shorter wait.
Models stay loaded on GPUs that are already hot, so a job skips loading weights and warming up and goes straight to generating.
Relative timeline — illustrative, not to scale.
Batching, concurrency, memory and scheduling are set per workload — the same model and output, fewer GPU-seconds.
The scheduler packs jobs onto GPUs back to back by model, resolution and length, so the same hardware finishes more work with less idle time.
Striped areas are idle GPU time. Illustrative.
Pillar 01 · Official partnerships
We work directly with ByteDance, Alibaba and MiniMax, so we get their lowest partner and channel pricing — on exactly the models they publish.
up to −46%
below the official list, live from our catalog
up to −51%
below the official list, live from our catalog
up to −49%
below the official list, live from our catalog
Run the numbers
Pick a model and how much you generate. Every price comes from our live catalog, next to the official list price.
Drag the slider — or pick a model — to try your own numbers.
Wan 3.0 at 1080p: 560 minutes of video per month costs $6,720 at the official list price and $3,310 on Cheaper Gen. — you save $3,410 a month.
Side by side
| Aspect | Each lab directly | Cheaper Gen. |
|---|---|---|
| Models | One lab per account and API | ByteDance, Alibaba and MiniMax in one API |
| Price | Official list price | Up to 51% below list |
| Routing & failover | Build and maintain it yourself | Automatic, on every request |
| Serving | The lab's standard API | Partner channels + our own tuned GPU pools |
| Billing | A contract and invoice per lab | One prepaid balance, pay per use |
Create an API key in a minute, or talk to us about volume and dedicated capacity.
