CheaperGen
NewSeedance 2.5 and Wan 3.0 are live

The best AI video & image models. For less.

Seedance, Wan, MiniMax and Seedream — create in a friendly Studio, or build with one simple API. Same official models, up to 93% below list price.

Start creatingGet an API key Talk to sales
  • Official models
  • Pay per second
  • Failed runs are free
  • Private by default
720p · 5 sdone
$1.08Official price $1.16−7%

Models & prices

Every model, below list price.

Official video and image models from ByteDance, Alibaba and MiniMax. Our price sits right next to the official one — no credits maths, no subscription needed.

Video models: modes, official list price, CheaperGen price and discount. Lowest tier, per second of video.
ModelPrice
Seedance 2.5
ByteDance · Text to Video · Image to Video · Reference to Video · Video Edit
from
$0.097/s
Official price $0.103−6%
Seedance 2.0 Fast
ByteDance · Text · Image · Reference → Video
from
$0.051/s
Official price $0.056−11%
MiniMax H3
MiniMax · Text · Image · Reference → Video
from
$0.057/s
Official price $0.08−29%
Kling 3.0
Kuaishou · Text · Image → Video
from
$0.078/s
Official price $0.084−8%
Veo 3.1 Fast
Google · Text · Image → Video
from
$0.074/s
Official price $0.08−8%
Wan 3.0
Alibaba · Text · Image · Reference → Video
from
$0.038/s
Official price $0.05−24%
MiniMax H3 Fast
MiniMax · Text · Image · Reference → Video
from
$0.024/s
Official price $0.04−41%
LTX-2.5
Lightricks · Text · Image → Video
from
$0.029/s
Official price $0.09−68%

Lowest tier shown, per second of video. Official list prices verified Oct 18, 2026.Every tier on the pricing page

  • Seedance
  • Seedream
  • Veo
  • Nano Banana
  • Wan
  • LTX
  • OpenMOSS

Explore

Meet the models. Make something.

Cinematic video, expressive motion, long takes and pixel-perfect images — pick a look and try it in seconds.

What the models can do: Text → Video, Image → Video, Native audio, Up to 4K, 30-second takes, Reference characters, Multi-image edits, Pay per second, Failed runs are free.

How we're cheaper

Same official models. Faster and cheaper.

Three things work together behind every price you see: direct partnerships with the labs, our own GPU inference for open-source models, and a router that sends each request to the fastest, most cost-effective healthy route.

Pillar 01

Official partnerships

We work directly with the labs that build the models, so we get their lowest partner and channel pricing — on the official models, untouched.

Pillar 02

Our own GPU inference

For open-source models we run inference ourselves: optimized kernels, warm GPU pools and per-model tuning make them several times faster and push the cost per second down further.

Pillar 03

Automatic routing

Every request is routed to the fastest, most cost-effective healthy channel, with live health checks and automatic failover.

Same official modelsFaster and cheaperOne API

Run the numbers

Pick a model and how much you make. Prices come straight from our live catalog, next to the official list price.

How it works
Model
Resolution

Drag the slider — or pick a model — to try your own numbers.

Wan 3.0 at 720p: 1,000 minutes of video per month costs $6,000 at the official list price and $4,542 on CheaperGen — you save $1,458 a month.

On average 18% below the official list across every model and tier. Inside our GPU inference

Powering products with millions of users. Read how they build on CheaperGen.

Case studies

Four ways in

Whether you make, ship or scale.

Same models, same honest prices — whichever way you like to work. And if you have capacity, sell it to us.

For creators

Create in the Studio

Pick a model, describe your shot, hit generate. No code, no subscription needed — you see the price before every clip.

  • Every model in one place
  • Pay per clip, from cents
  • Private by default — no public gallery
Open the Studio

For developers

Build with the API

One consistent API for every model. Submit a job, get a webhook. Or let your AI agent use it — connect over MCP.

  • Copy-paste JS, Python, cURL
  • Webhooks and async jobs
  • MCP for Claude Code, Cursor and more
Get an API key

For enterprise

Scale with us

Volume pricing below our public rates, dedicated capacity and a team that answers on Slack.

  • 99.9% uptime SLA
  • Invoicing and annual contracts
  • Data retention controls
Talk to sales

For suppliers

Sell us your capacity

GPU clouds, model labs and resellers: plug in spare GPU capacity or official-model credits, and our router keeps it busy with real traffic from creators, developers and enterprises.

Become a supplier

Developers

From API key to first video in two minutes.

One consistent API for every model. Submit a job, then poll or receive a webhook.

  1. 1
    Create an API key

    Sign in with Google or email and add credits as you need them.

  2. 2
    Call any model

    Same request shape for every model, documented on its page.

  3. 3
    Get results

    Poll the generation or receive a signed webhook when it's done.

Read the quickstart API reference
const API = "https://cheapergen.com/api/v1";
const headers = {
  Authorization: `Bearer ${process.env.CHEAPERGEN_API_KEY}`,
  "Content-Type": "application/json",
};

// 1. Submit the job
let job = await fetch(`${API}/run/bytedance/seedance-2.0-fast/text-to-video`, {
  method: "POST",
  headers,
  body: JSON.stringify({
    "prompt": "A slow dolly-in on a ceramicist shaping a vase at golden hour, dust in the light, shallow depth of field",
    "aspect_ratio": "16:9",
    "resolution": "720p",
    "duration": 5
  }),
}).then((r) => r.json());

// 2. Poll until it finishes (or pass webhook_url and skip polling)
while (job.status === "queued" || job.status === "running") {
  await new Promise((r) => setTimeout(r, 3000));
  job = await fetch(`${API}/generations/${job.id}`, { headers }).then((r) => r.json());
}

console.log(job.status === "succeeded" ? job.output.video.url : job.error);

AI agents

Use it from your AI agent.

Paste one sentence into Claude Code, ChatGPT, Codex, Cursor or any agent, and it sets itself up. Sign in once — no keys to paste. Then just ask for the video or image you want.

Try asking: “Animate this photo into a 6-second vertical clip with Wan 3.0.”

Paste this into your agent

Set up CheaperGen for me by following https://cheapergen.com/agents/setup.md

What your agent reads
Works with Claude Code, ChatGPT, Codex, Cursor and any other agent: it sets itself up, or walks you through the clicks. It asks before touching global settings and never sees your keys.
Setup for every agent

Why teams trust CheaperGen

FAQ

Questions, answered

Can't find what you need? Email us — a real person replies.

Yes. These are the original models from ByteDance, Alibaba, MiniMax and the other labs, served through official lab partnerships or our own GPU inference for open-source models. Same models, same quality, lower price. Accelerated variants, such as MiniMax H3 Fast, are labelled as such: a faster build of an open-weights model, listed as its own model next to the original.

Three ways. Direct partnerships with the official model labs give us their lowest channel pricing. And for open-source models, our own GPU inference — scheduling, warm pools and per-model tuning — brings the cost down further. Automatic routing sends every request to the fastest, most cost-effective healthy channel. We pass most of the saving on to you, and every model page shows our price next to the official list price.

No. The Studio lets you pick a model, write a prompt and generate in the browser — you see the exact price before every run. Or simply let your AI agent (Claude Code, Cursor, Claude and others) use CheaperGen through our MCP server: sign in once, no keys to paste.

Prepaid credits, pay-as-you-go. Top up with a card through Stripe, optionally turn on auto top-up, and you're charged per second of video or per image. No subscription is needed; optional monthly plans add credits every month, and Pro and Max lower prices. Failed generations are refunded automatically.

Yes — volume discounts, invoicing, dedicated capacity, higher rate limits, uptime SLAs and a dedicated support channel. Contact sales from the Enterprise page.

Yes. What you make is private to your account — there's no public gallery, and your prompts and creations are never used to train models. Studio creations stay until you delete them, and you can delete anything at any time. API request data is deleted automatically after 1 day by default, with configurable retention for teams. Payments go through Stripe.

Generate more. Spend less.

Add credits and make something today. No subscription needed, and failed runs never cost a cent.

Start creating Get an API key