For open-source models
Pay for the GPU,
not the idle time.
Load factor pricing ties your price for open-source models to n4n's real-time GPU load across its fleet — the lower the current load, the bigger your automatic discount, up to 80% off standard pricing.
Like gas fees, inverted
Crypto networks charge more when the network is congested; n4n does the opposite — low load means idle GPU capacity going unused, so your price drops instead of rising.
up to 80% off · automatic · no bidding
Right now, on n4n
The lower the load, the lower your price.
Llama 4 70B · list $0.35 / M tok → $0.16 / M tok right now
Illustrative — recalculated live from a simulated load cycle. Not real billing or telemetry data.
How it works
Automatic, every request.
01
We watch our own load
n4n tracks real-time GPU utilization across its fleet, continuously.
02
Idle capacity gets discounted
When load is low, open-source model pricing drops automatically — no action needed on your side.
03
You see it on every request
The discount that applied is shown in your usage log, per request, same as any other charge.
What this looks like
A few examples.
Llama 4 70B
typically 20–70% off
Mistral Large
typically 15–60% off
DeepSeek V4
typically 25–80% off
Qwen 3 72B
typically 20–65% off
Illustrative ranges — actual discount moves with real-time load and isn't fixed.
Start saving on open-source models.
Sign in, create a key, and every eligible request gets the discount automatically.
Open the console