aimodels.cheap
Live GPU spot price: /h · updated

AI video generation
at off-peak GPU prices.

Drop your prompts. We batch them and run them on spot GPUs when they are cheapest. Your videos come back within 24 hours, at roughly a tenth of what the APIs charge.

Get early access See live prices
MiniMax H3 · 768p · 24 h
/ video second
Full quality, 50 steps. APIs: $0.08
Draft mode · 24 h
/ video second
Turbo LoRA, 8 steps. 1,000 seconds of video for about a dollar.
10-second clip · 768p
per clip
APIs charge $0.80 for the same clip.

Live pricing indexed on spot

We run at cost. Prices are recomputed from the live Vast.ai RTX 5090 spot market and locked when you submit a job. Pay in USDT (BEP-20) and you get the raw price. Card payments via Stripe are coming, at +10 % to cover processing fees.

ModelWithin 24 hWithin 4 hWithin 1 hTypical API

All prices in USD per second of generated video. Failed or evicted jobs are never billed. Minimum top-up: $10 in USDT, $20 by card.

How it works

The same open-weight model the APIs run, on the same GPUs. We just refuse to pay for an idle one.

01 · QUEUE

Submit prompts

Text, image or reference video in, via API or dashboard. Pick a deadline: 24 h, 4 h or 1 h. The price is locked at submission.

02 · BATCH

We wait for the dip

A scheduler watches the spot market. When the queue is deep enough and GPUs are cheap, it rents them, loads the weights once and renders the whole batch back to back.

03 · DELIVER

Webhook or download

Each finished clip lands in your bucket or triggers your webhook. Jobs that miss their deadline are rerun on on-demand GPUs at our cost.

Why it is this cheap

Cost driverReal-time APIaimodels.cheap
GPU utilisationWarm GPUs waiting for requests, paid 24/7GPUs rented only while a batch is rendering
GPU priceOn-demand or reserved datacenter cardsInterruptible consumer RTX 5090 at the lowest bid of the day
Weight loadingAmortised per requestLoaded once per batch of dozens of clips
MarginWhatever the market accepts0.5 % on USDT during launch. Yes, half a percent.

FAQ

What quality do I get?

Full-quality MiniMax H3 at 50 denoising steps, 1344×768, 24 fps, with native stereo audio, up to 15 seconds per clip. "2K" is our 768p output upscaled locally, not the API's native 2K. Draft mode uses the official 8-step turbo LoRA: same model, faster, slightly softer motion.

Why does the price move?

Because our cost moves. The price is a fixed formula on top of the live RTX 5090 spot price: render seconds × GPU rate × 1.25 for cold starts and evictions × 1.005 margin. You always see the current number, and it is locked the moment you submit.

How do I pay?

USDT on BNB Chain (BEP-20), no account verification. Every account gets its own deposit address and the balance is credited within a minute. Card payments through Stripe are coming, at +10 % to cover fees and chargebacks. Credits are prepaid and never expire.

What if I need it faster than 24 h?

Pick the 4 h or 1 h tier. They cost more because we batch less and sometimes fall back to on-demand GPUs. If you need seconds, use a real-time API. We are the slow lane.

Is there an API?

Yes. Sign up, top up in USDT, submit jobs, get a webhook on completion. See the API docs.

Can I rent out my own GPU?

Not yet. At launch we rent from public spot markets. A supply side for RTX 4090 / 5090 owners is planned once demand exceeds what the spot market provides.

Early access

We are onboarding a first batch of users. Tell us roughly how many clips you generate per month and we will get back to you with an API key.

No spam. One email when your key is ready.