Model family	Resolution	Frames / rollout	Latency	Price / Frame	Status
DreamDojo 4 checkpoints · GR-1, AGIBOT, G1, YAM	480×640	49	2.6s	$0.0005	Live	Try model API

Dream Engines

Dream Engines — pricing

Dream Engines prices model APIs by generated frame. Each model publishes its own per-frame rate, and rollout prices are derived from that rate and the number of frames generated.

The current DreamDojo 2B family starts at $0.0005 per generated frame. A standard 49-frame rollout on those models costs $0.0245.

How to read the table

Unit	What it means
Price / frame	The metered API unit. Every successful generated frame decrements prepaid credits.
Price / rollout	A convenience estimate for the model's standard rollout length. DreamDojo rollouts are 49 frames today.
Price / 1M frames	A planning view for larger evaluation, augmentation, or MPC workloads.

Future open-world models may have different per-frame rates based on model size, resolution, rollout length, latency target, and hardware class. The API still reports the exact cost_usd for each completed request.

How billing works

Create or receive an API key with a prepaid credit balance.
Add credits with Stripe Checkout.
Run predictions through the Python SDK or API.
Each successful request deducts the exact per-frame cost from your balance.

If the engine fails to deliver a request, the charge is automatically refunded. If your balance is too low, the request is rejected before GPU work starts.

PYTHON

import dream
 
client = dream.Client()  # reads DREAM_API_KEY
 
print(client.billing.balance().balance_usd)
 
rollout = client.models.get("dreamdojo-2b-gr1").predict(
    start_frame=img,
    actions=acts,
)
 
session = client.billing.topup(amount_usd=25.00)
print("Open this checkout URL:", session.url)

What's included

Per-model frame pricing with no platform fee.
Prepaid credits that can be topped up as needed.
Automatic refunds for engine-side failures.
Per-request cost details surfaced in the API and SDK.
Rolling status and latency reporting at /v1/status.
Python SDK errors that clearly distinguish insufficient credits from engine failures.
Quality checks tracked with PSNR, SSIM, and LPIPS in docs/RESULTS.md.

Credits and limits

Limit	Default
Minimum top-up	$5
Maximum self-serve top-up	$10,000
Default refill rate	2 requests / second
Default burst	10 requests

Need a larger top-up, higher rate limit, invoice billing, or a dedicated rollout plan? Email hello@dreamengines.run and we'll configure your account.

Estimate your workload

The cost formula is:

generated frames × model price per frame

For example, a visual MPC workload on a $0.0005/frame model with 8 candidate rollouts per decision at 10 decisions per second for 1 hour:

1 hour × 10 decisions/sec × 8 candidates × 49 frames = 14,112,000 frames
14,112,000 frames × $0.0005 = $7,056

For smaller usage, 100 standard rollouts per day is 4,900 generated frames, or $2.45/day.

FAQ

Do credits expire?

No. Credits stay on your account until they are used.

Do I pay for failed requests?

No. Engine-side failures are automatically refunded.

Is there a free tier?

Not on self-serve pricing. If you're evaluating Dream Engine for a research or enterprise deployment, contact us and we can help scope a test budget.

Will every model cost the same?

No. DreamDojo 2B models share the same launch price today, but larger open-world models may carry higher per-frame rates when they require more expensive hardware, higher resolution, or multi-GPU serving.

Can I get committed-use pricing?

Yes. For larger workloads, procurement requirements, or dedicated support, email hello@dreamengines.run.

Model APIs