A machine you already own serves open models in the mesh, and you are paid per token it serves.
Your scenario
What could your node earn?
Start with your hardware and machine count. Fine-tune the workload in Advanced.
1 × 1 × 32 GB32 GB nominal aggregate memory
Allow room for runtime overhead and cache. Combining memory requires a suitable parallel profile and interconnect. This is a memory screen, not certified compatibility. Hardware specifications ↗
6 of 16 models meet the documented nominal memory target. All models remain visible. CUDA, ROCm and Apple layouts require their own serving profiles. Memory is checked per copy. Distributed layouts need a validated parallel profile.
Example workload: Qwen3.8-27B, 1 independent copy. Assumes 50 aggregate output tokens/s per copy and 4 concurrent requests per copy; illustrative, not a hardware benchmark.
Between 0 and 24 hours.
0–100% of online time spent on paid work.
Workload assumptions
Describe the load at which throughput was measured. This does not multiply the token rate.
Starts at an illustrative 50 tokens/s, not a measured hardware result. Replace it with sustained throughput for your workload. Enter the combined output of all concurrent requests for one copy, including time spent processing input.
The default of 3 is an assumption. Enter your workload’s ratio from 0 to 100. Throughput should reflect the time spent processing both input and output.
Estimated inference earnings
$262.05/ month
Before electricity, hardware, and other costs.
30 days · 24 hours online/day · 75% paid utilization
Per fully utilized hour
$0.49
Paid hours / month
540
Throughput assumption: 50 tokens/s total across 1 copy. Edit in Advanced → Workload assumptions.
Explore paid demand
Scenarios, not a traffic forecast. Select one to update the estimate.
Rates and calculation
Rates published · profile validation pending. Rate as of 2026-09-12
Input / 1M tokens
$0.36
Output / 1M tokens
$2.29
Host share
80%
tok/s × 3600 × blended rate × share
Blended rate = output price + input price × input/output ratio, converted to a per-token rate. Monthly earnings multiply the hourly estimate by online hours, 30 days, and paid utilization.
Utilization is the traffic we route to you, not a promise. Your console shows the real number. Being online and answering canaries do not themselves earn payment. This estimate covers inference only.
Earnings depend on demand, and the mesh is not open yet. This is arithmetic on published rates, not a promise.
shardio