Pricing
One price per unit, per template.
| Template | Unit | Price per unit | Needs | Availability |
|---|---|---|---|---|
| text/embeddings@1 | one batch of up to 1,000 texts of up to 512 tokens each (512k tokens), bge-base vectors | $0.001 | CPU | Available |
| video/transcode@1 | one video file of up to 60 s and 1,800 frames, 64x64 to 1280x720, anything ffmpeg reads, to H.265 (default) or AV1 MP4 | $0.004 | CPU | Available |
| audio/whisper-large-v3@1 | one audio file up to 10 minutes, transcript JSON | $0.004 | 12 GB GPU | Launching |
| image/flux-schnell@1 | one 1024x1024 image, 4 steps | $0.004 | 24 GB GPU | Launching |
| image/sdxl-turbo@1 | one image, 512x512 at 2 steps or no more work (up to 768 a side, 1 to 4 steps) | $0.0015 | 16 GB GPU | Launching |
| render/blender@1 | one frame of a .blend file, Cycles, PNG | $0.02 | 16 GB GPU | Launching |
Bring your own container
Your own container is quoted after a 20-unit calibration run (on up to two GPU classes) at three times the better measured cost per unit.
How the quote is madeCredits
New accounts start with $5 of credits. Top up by card in any amount from $5 to $10,000 (whole dollars in the dashboard, cents too from the CLI).
Start with $5 of free creditsPools
A job's pool says whose machines may run it: any that can (the default); yours, ours and other organizations' of trust 80 or more; ours alone; or yours alone. A unit costs the same in the first three, where fewer machines can mean a longer wait, and nothing when your own machine does it.
Choosing a poolHow prices are set
Every node in the fleet is benchmarked on every template. A template's price is the median cost per unit across GPU classes, times three, rounded. The fleet has not been measured yet: until the first measurement is published here with its date, the prices above are launch estimates. Work is placed by cost per unit, not by hourly price: a node takes a batch of units unless another free node that is asking for work would do them more than 10% cheaper.
Supplying a machine
A machine you supply earns the rate you set for the time each unit it completes for another organization took, and never more than 80% of the unit's price; a machine whose cost per unit is above 80% of a job's price is not offered that job.
The rates suggested for each GPU class, in US dollars an hour:
- a10
- $0.40
- a100
- $0.90
- a4000
- $0.12
- a5000
- $0.18
- a6000
- $0.40
- cpu
- $0.05
- h100
- $1.80
- l4
- $0.35
- l40s
- $0.80
- rtx3060
- $0.07
- rtx3060ti
- $0.08
- rtx3070
- $0.09
- rtx3070ti
- $0.10
- rtx3080
- $0.12
- rtx3080ti
- $0.14
- rtx3090
- $0.20
- rtx3090ti
- $0.22
- rtx4060
- $0.07
- rtx4060ti
- $0.09
- rtx4070
- $0.12
- rtx4070super
- $0.13
- rtx4070ti
- $0.15
- rtx4070tisuper
- $0.17
- rtx4080
- $0.20
- rtx4080super
- $0.21
- rtx4090
- $0.30
- rtx5070
- $0.14
- rtx5070ti
- $0.20
- rtx5080
- $0.26
- rtx5090
- $0.45
- rtx6000ada
- $0.60
- rtxpro6000
- $0.90
- t4
- $0.15
Any other class is suggested $0.0125 an hour for each GB of its VRAM, never less than the cpu rate. How earnings work